the daily brief

Cyber / Brief — 15 Sep 2026

State hackers had a busy weekend at the edge of the network: Russia's Sandworm was caught chaining flaws in Cisco's firewall-management software to resurrect its Cyclops Blink botnet on a new generation of Linux servers, while Chinese groups pressed a parallel campaign — one cluster phishing NGOs to plant a stealthy backdoor through a Chrome-and-Windows exploit chain, another, "Red Heron," ransacking self-hosted code repositories across six countries to steal source and burrow in. Europe's own cyber agency put words to the anxiety beneath it all: ENISA warned that frontier AI is collapsing the time between a flaw's discovery and its weaponisation from months toward minutes, and that the continent's human-paced defences must catch up. The politics of that reckoning turned sharp — Barack Obama called for AI safety measures in a pointed rebuke of President Trump, who dismissed the doomsaying even as chip stocks slid, while Beijing's state media branded Anthropic's slow-down proposal a "new Cold War playbook." And Brussels pressed ahead with rules of its own, moving to bar under-15s from opening social-media accounts under a new EU Kids Act, even as a Google DeepMind researcher warned on the way out that AI "may kill us all."

AI & Power

Obama Calls For AI Safety Measures, Rebuking Trump Approach — Bloomberg Politics
Why it matters: Barack Obama publicly calling for AI safety measures — and pointedly rebuking the Trump administration's approach — is the AI-slowdown reckoning becoming a partisan political fault line, a former president lending weight to the safety camp against a White House that dismisses the risks.
Former President Barack Obama called for AI safety measures, explicitly rebuking the Trump administration's dismissive approach to AI risk. The intervention matters because it elevates the AI-safety debate — already reshaped this fortnight by the frontier labs' slow-down call and Anthropic's threat disclosures — into a high-profile political contest: a former president aligning with the caution camp against a sitting administration that frames AI primarily as a race to win and dismisses 'doomsaying.' It sharpens the partisan divide over AI governance in the US (Democrats pressing for scrutiny, the administration prioritising the China race and opposing guardrails) and raises the political salience of the issue heading into the midterms. For Europe, watching the US oscillate between accelerationist and precautionary impulses, a prominent Democratic voice backing AI safety signals that the American governance debate is intensifying and polarising — and that the transatlantic gap on AI regulation may widen or narrow depending on which side of the US argument prevails, making the domestic American politics of AI safety a variable that shapes the global governance landscape the AI Act is part of.

Google DeepMind Staffer Says AI May ‘Kill Us All’ in Exit Post — Bloomberg Technology
Why it matters: A Google DeepMind staffer warning in an exit post that AI 'may kill us all' is insider alarm from the heart of a frontier lab — a departing researcher's stark public warning adding to the week's chorus of voices, from inside the industry, that the technology's risks are being outrun by its development.
A Google DeepMind staffer, in a departure post, warned that AI 'may kill us all,' a stark expression of existential concern from within one of the world's leading AI labs. The exit warning matters as another instance of insider dissent — following the Anthropic resignation over the industry 'gambling with our lives' and the frontier CEOs' own slow-down call — signalling that serious safety-and-existential worry is being voiced by the researchers closest to the technology, not just outside critics. Whatever one makes of the specific claim, the pattern of public alarm from within the labs (DeepMind, Anthropic) lends weight to the argument that capability is outpacing safety and control, and it feeds the broader reckoning now spilling into markets and politics. For the governance debate — and for Europe's systemic-risk framing under the AI Act — the accumulation of insider warnings is a meaningful signal that the people building frontier AI are not uniformly confident it is safe, a data point that strengthens the case for external oversight where internal restraint may not suffice.

Trump Opposes AI Guardrails Amid Chip Selloff — Bloomberg Technology
Why it matters: President Trump digging in against AI guardrails even as chip stocks sell off on safety fears is the accelerationist case meeting the market's anxiety head-on — a White House betting that winning the AI race outweighs the risks the industry itself is now warning about.
President Trump reaffirmed his opposition to AI guardrails amid a chip-stock selloff driven by AI-safety fears, doubling down on an accelerationist, competition-first posture even as markets wobbled on the industry's slow-down warnings. The stance matters because it sets the US federal position in direct opposition to both the frontier labs' call for restraint and the market's evident anxiety: framing AI regulation as a threat to winning the race against China, the administration resists the binding constraints that safety advocates (and now a chorus including Obama) urge. It crystallises the core tension of the moment — safety versus speed, caution versus competition — now playing out across markets, politics and the labs themselves. For Europe, a US administration explicitly opposing AI guardrails while Brussels enforces the AI Act widens the transatlantic governance gap and increases the pressure on the EU as the world's most assertive AI regulator; and the market reaction underscores that the safety debate now carries real financial weight, making the politics of guardrails a matter of investor as well as public concern.

Hugging Face Says Good Ethics Can Drive AI Innovation — Bloomberg Technology
Why it matters: Hugging Face arguing that good ethics can drive AI innovation — not just constrain it — is a constructive counter-narrative to the safety-versus-speed framing, the open-model hub insisting that responsible development and progress can reinforce rather than oppose each other.
Hugging Face — the central hub of the open-model ecosystem — argued that good ethics can drive AI innovation, offering a counter to the framing of safety and progress as opposites. The argument matters as a constructive intervention in a debate that has hardened into 'slow down for safety' versus 'accelerate to win': Hugging Face's position is that ethical, responsible and safe development is not merely a brake but can be a driver of better, more trustworthy and more durable innovation, and (via its scientists' warnings on agentic-AI safety) that addressing risks is part of building good technology. It resonates with Europe's own thesis — embodied in the AI Act — that rights-respecting, safe AI can be a competitive advantage rather than a handicap. For the governance debate, Hugging Face's framing is a reminder that the safety-versus-speed dichotomy is not the only way to see the choice, and that a credible middle path — responsible innovation — is what much of the European approach aspires to, even as the US argument polarises between acceleration and alarm.

EU & Technology

ENISA: Frontier AI Is Changing the Speed of Cyberattacks. Europe Needs to Catch Up — Security Affairs
Why it matters: Europe's cybersecurity agency warning that frontier AI is collapsing the time between a flaw's discovery and its weaponisation from months toward minutes — and that the continent's human-paced defences must catch up — is the EU institutionalising the fortnight's central fear, and putting a number on it.
ENISA, the EU cybersecurity agency, warned in its 'Cybersecurity in the Frontier AI Era' assessment that advanced AI is compressing parts of the attack chain — reconnaissance, vulnerability discovery, exploitation, lateral movement, data theft — and that the gap between discovering a flaw and weaponising it could shrink from months to minutes (ENISA cites vulnerability-weaponisation windows compressed toward ~15 minutes, and one organisation's vulnerability-report volume rising from ~80 in Q1 2025 to about 500 per day once frontier-AI tools were in use). The warning matters because it is the EU's own agency formalising, with data, the 'AI compresses the exploit timeline' thread the fortnight has repeatedly evidenced (the Anthropic report, the machine-speed intrusions), and concluding that traditional human-managed patch cycles risk obsolescence while legacy and end-of-life systems become more exposed to AI-assisted discovery. It aligns with the European Commission's Action Plan on AI-era cybersecurity, and for European organisations under NIS2 it is an authoritative signal that defence must move toward automation and speed. ENISA's message — Europe needs to catch up — is the continent's security establishment naming the AI-and-cyber challenge as an urgent, structural one, not a future hypothetical.

EU to limit access to social media before age of 15 — Cybersecurity and Data Protection – POLITICO
Why it matters: The EU moving to bar under-15s from opening social-media accounts — a tiered 'EU Kids Act' with age-verification duties on Meta, TikTok and the rest — is Brussels making one of its boldest interventions yet in how children live online, with sweeping implications for platform design and age assurance across the bloc.
The European Commission is set to unveil an 'EU Kids Act' establishing a minimum age of 15 for children to open their own social-media accounts (with parental permission), part of a tiered system: under-3s barred from social media and other high-risk services, 3-13-year-olds limited to child-friendly services under adult supervision, under-15s needing parental consent, and 15-18s restricted to 'safe by design' versions — with platforms required to verify the age of any new user. Announced by von der Leyen this week, it matters as a landmark, bloc-wide child-online-safety intervention that would reshape how platforms like Meta and TikTok design and gate their services for minors across Europe, and it lands amid the global age-assurance wave (the UK's device-level protections, national schemes). The mandate's requirement to verify users' ages is consequential and contested: it advances child protection but raises the same privacy-and-implementation questions (age verification without eroding anonymity, avoiding surveillance-by-default) that European digital-rights advocates keep flagging. For the EU's digital rulebook, the Kids Act is a bold assertion of the bloc's willingness to regulate platform design in the name of child safety — and a test of whether it can do so in a rights-respecting, privacy-preserving way.

Dutch Train Disruption Looks Like Sabotage, Government Says — Bloomberg Technology
Why it matters: The Dutch government saying a train disruption looks like sabotage is another European state naming an infrastructure incident as a likely deliberate attack — a marker of how thoroughly the expectation of hybrid aggression now frames the reading of disruptions to the continent's transport arteries.
The Dutch government said a train disruption appears to have been sabotage, adding the Netherlands to the growing list of European states attributing infrastructure incidents to likely deliberate attack. The assessment matters because rail and transport infrastructure has become a focal point of Europe's hybrid-threat concern (the Leipzig sabotage, the Kyiv-Warsaw train strike, the RUSI analysis that transport underpins NATO's defences), and a national government publicly characterising a disruption as sabotage reflects both the reality of the threat and the now-reflexive framing of ambiguous incidents through that lens. It comes with a caveat the fortnight itself surfaced — that not every incident is Russian sabotage, and threat inflation carries its own risks — so the Dutch assessment underscores the importance of evidence-based attribution even as it signals genuine concern. For European security and critical-infrastructure protection, the episode is another data point in the pattern of attacks (or suspected attacks) on transport, and a reminder that the resilience and security of the continent's rail and logistics networks — physical and digital — is now a front-line concern that governments are treating with increasing seriousness.

US & Technology

Trump’s AI Defense Defies Voter Unease Heading Into Midterms — Bloomberg Technology
Why it matters: President Trump's staunch defence of AI against growing voter unease is the accelerationist bet meeting the electorate — a White House wagering that championing the AI race will play better than caution, even as public anxiety about the technology rises ahead of the midterms.
President Trump's vigorous defence of AI — dismissing safety concerns and championing the race against China — is running up against growing voter unease about the technology, a tension that adds to his party's risks heading into the midterms. The dynamic matters because it places the AI-safety debate squarely into electoral politics: as public anxiety about AI rises (job displacement, safety, the mainstreaming of AI-risk fears) and prominent voices including Obama call for guardrails, the administration's accelerationist, anti-guardrail posture becomes a political gamble — betting that voters reward winning the AI race over managing its risks. It connects to the broader US political contest over AI (the Democratic push for scrutiny, the calls for safety measures) and to the question of whether public sentiment will force a shift in the permissive US regulatory stance. For Europe and the transatlantic governance picture, the electoral salience of AI in the US is consequential: if voter unease translates into political pressure, the American approach to AI regulation could move closer to Europe's precautionary stance — or, if the accelerationist bet pays off politically, the transatlantic gap could persist, making the US midterms a variable in the global trajectory of AI governance.

DOJ Says Millions in Iran Oil Proceeds Laundered on Binance — Bloomberg Technology
Why it matters: The US Justice Department alleging that millions in Iranian oil proceeds were laundered through Binance is the crypto-and-sanctions-evasion nexus back in the spotlight — a reminder that the largest crypto exchange remains a focus of enforcement over the movement of illicit and sanctioned funds.
The US Department of Justice alleged that millions of dollars in Iranian oil proceeds were laundered through the Binance cryptocurrency exchange, tying the platform to sanctions evasion and illicit-finance flows. The allegation matters because it reinforces the enforcement focus on cryptocurrency's role in sanctions evasion and money laundering — Binance has faced prior major US enforcement over compliance failures, and the movement of sanctioned Iranian oil revenue through crypto rails is exactly the illicit-finance activity regulators and prosecutors target. It connects to the broader theme of cryptocurrency as an instrument of sanctions circumvention (by Iran, North Korea, Russia and others) and to the scam-and-laundering economy the fortnight has tracked (the FinCEN scam-centre alert, the Xinbi marketplace takedown). For Europe — aligned with the US on Iran sanctions and pursuing its own crypto-and-AML regime (MiCA, the AML package) — the case is a reminder that crypto exchanges remain a critical control point in the fight against sanctions evasion and illicit finance, and that enforcement against the laundering of sanctioned funds through major platforms is an ongoing priority with transatlantic dimensions.

China & Technology

China Calls Amodei’s AI Proposal a New Cold War Playbook — Security Affairs
Why it matters: Beijing branding Anthropic CEO Dario Amodei's AI slow-down proposal a 'new Cold War playbook' is China reading the West's safety turn as a containment strategy in disguise — a reminder that the AI-safety debate is inseparable from the geopolitical rivalry, and that calls for restraint are heard very differently across the divide.
Chinese state media dismissed Anthropic CEO Dario Amodei's call to slow AI development as 'self-serving' and framed his proposal as a 'new Cold War playbook,' casting the Western safety turn as a strategy to constrain China's AI progress under the guise of caution. The response matters because it reveals how differently the AI-safety debate lands across the geopolitical divide: where the frontier labs and Western commentators frame the slow-down as risk management, Beijing reads it as an attempt to freeze the current hierarchy and deny China the chance to catch up — a suspicion sharpened by the recent US export controls, the distillation advisory naming Chinese firms, and the framing of AI as a race to win. It connects to China's own AI-risk warning (its state-security minister's caution) while showing that Beijing rejects the specific Western proposal as competitively motivated. For the prospect of any coordinated, cross-border restraint on frontier AI, China's dismissal is a significant obstacle: a genuine global slowdown requires the participation of all major powers, and Beijing's reading of the proposal as a Cold War manoeuvre suggests the geopolitical rivalry will keep undermining collective safety efforts, leaving each side racing partly out of distrust of the other.

Chinese AI Chip Darling Biren Said to Mull $1 Billion Share Sale — Bloomberg Technology
Why it matters: Chinese AI-chip champion Biren reportedly weighing a $1-billion share sale is more capital flowing into China's drive for domestic AI silicon — the market and the state together funding the homegrown alternatives to Nvidia that Beijing's self-sufficiency push depends on.
Biren Technology, a prominent Chinese AI-chip designer, is reportedly considering a $1-billion share sale to fund its expansion, adding to the wave of capital pouring into China's domestic AI-hardware sector (following Enflame's Shanghai debut and Z.AI's multibillion-dollar raise). The fundraising matters because China's AI ambitions hinge on building capable homegrown accelerators to substitute for restricted US hardware, and continued market enthusiasm and capital availability for domestic chip champions like Biren reflect both national-strategic backing and investor conviction in the self-sufficiency story. It fits the fortnight's China-hardware thread — Huawei's 'un-American' chip, the domestic-accelerator mega-clusters, the drive to stretch compute — of a Chinese ecosystem marshalling capital and capability to close the gap that US export controls are meant to preserve. For Europe and the West, the steady flow of large sums into Chinese AI chipmakers is a reminder that Beijing's semiconductor self-sufficiency drive is well-resourced and accelerating, and that the durability of the West's hardware advantage depends on China's indigenous progress — which capital raises of this scale are explicitly funding.

China’s military research absorbs HK, Macau. Track the links with ASPI’s updated China Defence Universities Tracker — The Strategist
Why it matters: New tracking showing China's military research apparatus absorbing Hong Kong and Macau universities is the quiet militarisation of the region's academic-and-technology base — research institutions being drawn into the PLA's innovation pipeline, with implications for technology transfer and the risks of collaboration.
An Australian Strategic Policy Institute update to its China Defence Universities Tracker documents how China's military research apparatus is absorbing Hong Kong and Macau institutions, tracking the growing links between the region's universities and the PLA's defence-innovation ecosystem. The finding matters because it maps the extension of China's military-civil fusion into Hong Kong and Macau — universities that international partners have long collaborated with on the assumption of academic openness are increasingly tied to defence research, raising the risk that collaboration, talent and technology flow into China's military modernisation. It carries direct implications for research security: European and other institutions partnering with Hong Kong and Macau universities may be inadvertently contributing to PLA-linked research, a concern that has driven tightening scrutiny of academic collaboration and export controls on dual-use knowledge. For Europe — grappling with how to protect its research base and sensitive technology from illicit transfer (the Belgian chip-espionage case, the broader economic-security turn) — the absorption of Hong Kong and Macau research into China's military pipeline is a reminder that research-security and collaboration-vetting must account for the reach of China's military-civil fusion, and that the openness of academic exchange is being exploited as a technology-transfer channel.

Threat Intelligence (CTI)

[P1] 'Sandworm' Chains Cisco Vulnerabilities to Deploy Cyclops Blink — darkreading
Why it matters: Russia's Sandworm was caught chaining flaws in Cisco's firewall-management software to resurrect its Cyclops Blink botnet — rebuilt for modern Linux servers with new scanning and packet-sniffing powers — the GRU's most destructive cyber unit turning enterprise security infrastructure into the backbone of a revived espionage-and-attack network.
Sophos attributed, with high confidence to Russia-nexus actors and moderate confidence to Sandworm (tracked as Iron Viking), a new campaign chaining two Cisco Secure Firewall Management Center (FMC) flaws to deploy an upgraded Cyclops Blink botnet. The chain uses CVE-2026-20079 (a maximum-severity, CVSS 10.0 authentication bypass letting an unauthenticated remote attacker run code and gain root on the FMC OS) together with CVE-2026-20316 (low-privileged-account login); attackers first drop a Netcat-based reverse shell and proxy tool on vulnerable FMC systems, then deploy the new Cyclops Blink variant. The rebuilt malware runs on 64-bit x86-64 Linux (vs the original 32-bit PowerPC), uses generic Linux persistence rather than firmware modification, and adds network scanning, packet capture and expanded credential-and-system-data collection. Cisco Talos disclosed the broader FMC exploitation on 9 September, warning attackers gained access, deployed reverse shells and proxy tooling, stole device data and installed Cyclops Blink.
severity critical (CVSS 10.0) · exploited in the wild · CVE-2026-20079 · EU: NIS2, DORA, CER Directive · actor Sandworm / Iron Viking (Russia GRU; moderate confidence) (60%), escalation

[P2] China-Linked Hackers Exploit Chrome-Windows Zero-Day Chain to Deploy GRIMWEDGE — The Hacker News
Why it matters: A Chinese threat cluster spear-phished NGOs with a Chrome-and-Windows exploit chain that slipped through a 'patch gap' to compromise fully-updated machines — planting a stealthy JavaScript backdoor called GRIMWEDGE by hiding the attack behind a fake charity donation page.
Volexity attributed to a Chinese threat cluster it tracks as UTA0560 a spear-phishing campaign (observed 1 September 2026) that exploited a Chrome-and-Windows exploit chain to deploy a JavaScript backdoor named GRIMWEDGE against multiple NGOs. The intrusion began with spear-phishing emails linking to legitimate US university websites vulnerable to reflected cross-site scripting, abused as redirectors to attacker-controlled landing pages; victims saw a convincing donation-form image while a hidden iframe launched the exploit chain: CVE-2026-85046 (arbitrary read/write in Chrome's V8) -> CVE-2026-87491 (V8 sandbox escape) -> CVE-2026-85880 (code injection into the Chrome process for RCE). The attack exploited a 'patch gap' - CVE-2026-85046 was fixed upstream in Chromium before the attacks but had not yet reached stable Chrome, so fully-updated-looking users were still exposed. GRIMWEDGE (a JScript backdoor) can collect system info, list directories and processes, read/delete files, execute commands, terminate processes and fetch further payloads. It is part of the broader multi-cluster Chinese exploitation of this chain (the 'BlueMoon'/patch-gap campaign).
severity high · exploited in the wild · CVE-2026-85046 · EU: NIS2 · actor UTA0560 (China-nexus; Volexity) (60%), escalation

[P2] Red Heron Exploits Gitea RCE to Compromise 13 Organizations Across Six Countries — The Hacker News
Why it matters: A Chinese-speaking group dubbed Red Heron turned a public exploit for a flaw in Gitea — a self-hosted rival to GitHub — into an automated weapon, ransacking source-code repositories at organisations across six countries to steal code and burrow in with a custom Linux implant.
A Chinese-speaking threat actor tracked as Red Heron weaponised CVE-2026-60004, a critical (CVSS 9.8) remote-code-execution flaw in Gitea (a self-hosted source-code-management platform), affecting versions 1.17 through 1.27.0 and patched in 1.27.1 on 27 July 2026. Within days of the July disclosure, Red Heron transformed public proof-of-concept code into an automated Python framework (from 29 July) that registers accounts, exploits vulnerable servers, steals repositories and removes selected traces; it scanned 1,386 Gitea instances across seven countries (plus a separate dataset of 477 Taiwan-based systems) and compromised organisations across six countries (US, Taiwan, Canada, Argentina, Qatar, Sri Lanka). The activity progressed from source-code theft to persistent access, credential collection and lateral movement, and analysis of a staging server uncovered a C++ Linux implant dubbed JITTERLY supporting 30+ post-exploitation commands (shell execution, file transfer, process termination, network tunnelling, interactive terminal, internal pivoting).
severity high (CVSS 9.8) · exploited in the wild · CVE-2026-60004 · EU: NIS2, CRA · actor Red Heron (Chinese-speaking) (60%), escalation

[P2] 3BB Attacker Used MeshCentral Backdoor for Root Access, Targeted Subscriber Credentials — The Hacker News
Why it matters: An intruder lurking inside one of Thailand's largest broadband providers held root control of internal servers using a legitimate remote-management tool, and had built scripts to siphon the databases holding millions of subscribers' login credentials — a telecom compromise that abused trusted IT software to hide in plain sight.
Threat-intelligence firm Hunt.io detailed an intrusion inside 3BB, one of Thailand's largest broadband providers, in which an attacker maintained remote control of internal machines using MeshCentral, a legitimate free remote-management tool that IT teams use to administer computers. Hunt.io uncovered the operation by examining an internet-exposed server the attacker had left open (captured 3 June 2026 while the operation was live), which held the attacker's tools and a list of machines already under their control. A recovered file showed the attacker gaining full root control of an internal server and installing MeshCentral for persistence; scripts on the server were built to copy out 3BB's RADIUS databases - the systems storing the login credentials broadband customers use to get online. The evidence shows the RADIUS databases were targeted (not that data was confirmed exfiltrated), alongside scripts to expand access, gather credentials and probe internal systems.
severity high · exploited in the wild · EU: NIS2

[P3] Hundreds of fake government websites target users in Central Asia — The Record from Recorded Future News
Why it matters: Scammers stood up hundreds of fake government and news websites across Central Asia, dangling bogus state cash payments to harvest people's contact details — a mass fraud campaign trading on citizens' trust in government social programs to reel victims in.
Researchers at cybersecurity firm F6 identified more than 360 fraudulent domains in a campaign creating fake government and news websites to target people in Uzbekistan, Belarus and Tajikistan with bogus offers of cash payments or passive income. The sites exploit trust in government social programs - telling visitors they are eligible for financial assistance or can earn passive income through government-backed schemes (for example, promising Uzbek residents weekly payments of ~15 million sum / about $1,300) - and are designed to collect victims' contact details (initially just name and phone number), which the scammers then use to target them via phone or email to steal money, personal information, or gain access to their devices. It is a large-scale social-engineering-and-fraud operation abusing government-brand trust.
severity medium · exploited in the wild · EU: GDPR

Defence & National Security

NATO Shoots Down Drone That Entered Lithuanian Airspace — Bloomberg Politics
Why it matters: NATO shooting down a drone that entered Lithuanian airspace is the alliance's front-line air defences engaging an incursion on its eastern edge — a live instance of the drone-and-airspace threat that has become a defining feature of European security, met this time with force.
NATO forces shot down a drone that had entered Lithuanian airspace, an active engagement on the alliance's eastern flank amid a wave of drone incursions and airspace violations across Europe. The incident matters because unmanned-aircraft incursions — whether reconnaissance, provocation, spillover from the war in Ukraine, or deliberate testing of air defences — have become a persistent and destabilising feature of European security, and NATO downing a drone over a member state's territory is a concrete assertion of the alliance's willingness to defend its airspace. It connects to the fortnight's broader European-security and hybrid-threat thread (the sabotage incidents, the drone factory and airspace concerns, Denmark's report of a Russian warship firing flares at a helicopter) and to the challenge of countering cheap, proliferating drones with appropriate and proportionate defences. For European defence, the Lithuanian engagement underscores that airspace security on the eastern flank is now an operational reality requiring counter-drone capability and clear rules of engagement, and it is a marker of the heightened tension and the readiness to respond that increasingly define NATO's posture toward incursions on its frontier.

Japan’s record defence budget request is really about AI — The Strategist
Why it matters: The argument that Japan's record defence-budget request is 'really about AI' reframes the headline spending figure as a bet on the technology reshaping warfare — a major military power directing its rearmament toward the AI, autonomy and data capabilities that will define future conflict.
An Australian Strategic Policy Institute analysis argues that Japan's record defence-budget request is fundamentally about AI — that beneath the headline spending increase lies a strategic bet on the artificial-intelligence, autonomy, data and advanced-technology capabilities that are reshaping military power. The framing matters because it captures a broader shift in defence spending among advanced militaries: the decisive investments are increasingly in AI-enabled systems (autonomy, sensing, decision support, cyber) rather than only traditional platforms, reflecting the recognition — evident from Ukraine to the Indo-Pacific — that AI and autonomy are becoming central to warfighting. Japan directing its rearmament toward these capabilities, amid US pressure to spend more and rising regional tension with China, is a significant marker of how the AI-military nexus is driving defence priorities. For European militaries pursuing their own rearmament and drawing the same lessons, Japan's AI-oriented budget is a benchmark: credible modern defence increasingly means investing in the AI, autonomy and data capabilities that determine advantage, and the headline spending figures matter less than where the money goes — a reorientation Europe's own defence build-out is grappling with.

Denmark Says Russian Warship Fired Flares at Helicopter — Bloomberg Politics
Why it matters: Denmark reporting that a Russian warship fired flares at one of its helicopters is another sharp-edged encounter in the Baltic — a physical confrontation that fits the pattern of Russian provocation Europe's northern flank is increasingly absorbing.
Denmark said a Russian warship fired flares at a Danish military helicopter, a provocative encounter in or near the Baltic that adds to the tally of Russian confrontational behaviour toward NATO forces and infrastructure in the region. The incident matters as another marker of the heightened tension and grey-zone provocation on Europe's northern and eastern flanks — alongside airspace incursions (the Lithuanian drone), suspected sabotage (the Dutch train, Baltic cable and pipeline cases), and the broader pattern of Russian pressure that European governments increasingly treat as a coordinated campaign. Physical confrontations between Russian and NATO-member forces raise the risk of escalation and miscalculation, and they underscore the militarisation of the Baltic as a zone of contest. For European security, the flares incident is a reminder that the confrontation with Russia is not only cyber and hybrid but increasingly kinetic and direct at the military-encounter level, and that managing the risk of escalation while deterring provocation is a live challenge for NATO's members on the front line — part of the same deteriorating security environment driving the continent's rearmament and resilience efforts.

Digital Sovereignty & Identity

UK.gov begins killing off passwords for 23 million users — www.theregister.com - Articles
Why it matters: The UK government beginning to replace passwords with passkeys for 23 million users of its digital services is one of the largest state-scale moves yet toward passwordless authentication — a concrete step to blunt the credential theft and phishing that drive so many breaches.
The UK government has begun phasing out passwords in favour of passkeys for the roughly 23 million users of its GOV.UK/One Login digital services, one of the largest government-scale deployments of passwordless authentication. The move matters because passwords are the root of a huge share of breaches — phishing, credential theft, reuse, the vishing-and-MFA-bypass campaigns the fortnight has tracked (ShinyHunters, passkey-themed social engineering) — and passkeys (phishing-resistant, hardware-or-device-bound credentials) meaningfully raise the bar against exactly those attacks. A national government moving 23 million citizens toward passkeys is a significant real-world validation of the passwordless transition and a model for large-scale adoption, though it also concentrates the importance of the account-recovery and enrolment flows (which attackers increasingly target as MFA's weakest link). For European digital-identity and public-service security, the UK's passkey rollout is a notable marker of the shift away from passwords at scale — a phishing-resistance gain that, done well, could reduce the credential-driven breaches that plague the public sector, provided the recovery and fallback mechanisms are hardened against the social-engineering that has become the go-to bypass.

Piero Cipollone: The future of euro cash: trusted today, designed for tomorrow — ECB - European Central Bank
Why it matters: An ECB board member setting out the future of euro cash — trusted today, designed for tomorrow — is the central bank tending to physical money even as it builds the digital euro, a reminder that Europe's monetary sovereignty spans both the cash in people's pockets and the CBDC on the drawing board.
ECB Executive Board member Piero Cipollone set out a vision for the future of euro cash — 'trusted today, designed for tomorrow' — as the central bank simultaneously advances the digital euro and considers the evolving role of physical currency. The framing matters because Europe's monetary sovereignty and the resilience of its payment system depend on both pillars: cash remains a trusted, inclusive, offline-capable and privacy-preserving means of payment (and a fallback when digital systems fail), while the digital euro is being developed to secure European payment autonomy against dependence on non-European (largely US) card networks and private stablecoins. Cipollone's emphasis on cash's continued role, alongside the ECB's digital-euro work and Lagarde's framing of 'a new age of capital: growth, sovereignty and AI,' reflects a considered approach to monetary sovereignty that spans physical and digital. For Europe's digital-sovereignty agenda — where control of payment infrastructure is a strategic concern — the parallel attention to resilient cash and a sovereign digital euro is a reminder that the continent's monetary autonomy is being defended on both fronts, and that the design choices around the digital euro (privacy, offline capability, inclusion) will determine whether it extends rather than erodes the trust cash embodies.

EU lets 9 Schengen countries delay full EES biometric checks — Biometric Update
Why it matters: The EU allowing nine Schengen countries to delay full rollout of the Entry/Exit System's biometric checks is the messy reality of deploying continent-scale border-identity infrastructure — an ambitious biometric regime meeting the operational friction of readiness, capacity and queues at the frontier.
The EU has permitted nine Schengen countries to delay the full implementation of biometric checks under the Entry/Exit System (EES), the bloc's new automated system for registering non-EU travellers' biometric and travel data at external borders. The delay matters because the EES is a major piece of European border-identity infrastructure — capturing fingerprints and facial images of arriving travellers to replace passport stamping and strengthen border control — and its phased, flexible rollout reflects the operational challenges (readiness, capacity, technology, the risk of long queues) of deploying biometric systems at continent scale. It connects to Europe's broader digital-identity-and-biometric build-out (the EUDI wallets, the many national schemes) and to the tension between the security-and-efficiency goals of biometric borders and the practical and privacy considerations of implementing them. For European digital sovereignty and identity, the EES rollout — even in its delayed, phased form — is a consequential expansion of biometric data collection at the borders, and its bumpy deployment is a reminder that continent-scale identity infrastructure is as much an operational-and-logistical challenge as a technical or policy one, with the privacy and proportionality of large-scale biometric collection an ongoing concern.

Cybersecurity & Threats

[P1] Cisco Secure Email Gateway Flaw Exploited in the Wild, Enables Root Command Execution — The Hacker News
Why it matters: Attackers are exploiting a maximum-severity zero-day in Cisco's Secure Email Gateway — simply sending a crafted email seizes root control of the appliance guarding an organisation's mail, no login required — a critical, actively-abused flaw in a device at the perimeter of enterprise communications.
Cisco issued an urgent alert for CVE-2026-76461, a critical (CVSS 9.8) zero-day in its Secure Email Gateway appliances (Cisco AsyncOS Software) that is being actively exploited in the wild. A parsing flaw with insufficient input sanitisation lets an unauthenticated attacker embed malicious input in an email that, when processed by the appliance, triggers command injection and grants arbitrary command execution with root privileges — no credentials required. It affects both physical and virtual Secure Email Gateway appliances regardless of configuration, and attackers have weaponised it in the wild through September 2026. Cisco fixed it in AsyncOS 16.5.0-780 (plus earlier branches 16.0.4-3021 and 15.5.5-0141); there are no practical workarounds, making immediate patching imperative.
severity critical (CVSS 9.8) · exploited in the wild · CVE-2026-76461 · EU: NIS2, DORA, CER Directive

[P2] New DDRop Attack Breaks Intel TDX and AMD SEV-SNP Confidential Computing — The Hacker News
Why it matters: A new hardware attack called DDRop breaks the confidential-computing protections that cloud providers sell to shield sensitive workloads — silently dropping writes to encrypted memory so the processor keeps reading stale data — the first to defeat the integrity of an up-to-date Intel TDX system, not just read from it.
Researchers disclosed DDRop, an active interposer attack that breaks the memory-integrity protections of Intel TDX and AMD SEV-SNP (and Scalable SGX) confidential-computing platforms. A small board of switches sits on the DDR5 memory bus at full speed and, to 'drop' a write, forces an error on the command bus and then cuts the wire the memory module uses to report that error — so the module silently discards the command and the processor is never told, continuing to read old encrypted data. It is the first active interposer attack against DDR5 in current cloud servers and the first to break the integrity of an up-to-date Intel TDX system rather than merely read data. On AMD SEV-SNP, dropping writes during page-relocation let researchers copy one victim page into another; DDRop can expose victim memory, toggle debug mode and forge attestation on TDX. All three technologies encrypt memory without the freshness check DDRop exploits. It requires physical access (the interposer hardware); older Intel Client SGX (with an integrity tree, now retired) is unaffected.
severity high · EU: NIS2, DORA, CRA

[P2] Human Attacker Hits Machine-Speed Exploitation of Marimo RCE — Infosecurity Magazine
Why it matters: A human attacker — using no AI at all — moved from a vulnerable Marimo notebook to an SSH bastion host in eight seconds and looted cloud secrets over a nine-hour, hand-typed session, a striking counterpoint to the week's AI-speed narrative: the machine-speed threat can be human too.
Sysdig detailed a real-world exploitation of CVE-2026-39987, a pre-authentication remote-code-execution flaw in the Marimo notebook platform, notable because a human operator — showing no sign of using an LLM at any stage — achieved 'machine speed.' The flaw is in Marimo's terminal WebSocket endpoint, which skipped the authentication check applied to its other WebSocket endpoints, so any client connecting to it received an interactive shell as the Marimo process user with no credentials. The attacker moved from the vulnerable notebook to an SSH bastion host in about eight seconds, and over a roughly nine-hour session issued 850+ interactive commands, ran a full credential-pivot chain (unauthenticated WebSocket terminal -> AWS Secrets Manager call using harvested credentials -> SSH to a bastion host with the retrieved private key), used no recognisable public offensive tooling, and hand-rolled scripts in-session. CVE-2026-39987 affects Marimo up to 0.20.4, is fixed in 0.23.0, and has been on CISA's KEV catalogue for months (federal remediation deadline 7 May 2026).
severity high · exploited in the wild · CVE-2026-39987 · EU: NIS2

[P3] Malicious Twitch Browser Extension Leaks OAuth Tokens From Nearly 31,000 Users — The Hacker News
Why it matters: A popular Twitch browser extension available in the official Chrome and Firefox stores was quietly siphoning the login tokens of nearly 31,000 users to a Russian bot service — tokens that let anyone holding them take over a Twitch account without its password or two-factor code.
Researchers (Socket) found that a cross-store browser extension, 'Twitch Enhanced Viewer | JeetBot' — available in the official Chrome and Firefox stores — leaked the Twitch OAuth session tokens of nearly 31,000 users to proxy servers operated by JeetBot, a Russian commercial bot service. To deliver ad-free, region-unlocked video, the extension routed Twitch's video-playlist requests through operator-controlled proxies, and (from version 4.8, January 2026) POSTed captured tokens to a '/set-token' endpoint on JeetBot infrastructure. The leaked OAuth tokens let anyone who holds them act on affected Twitch accounts without the password or two-factor authentication. The version history was suspicious (jumping from 7.2.6 in April to 85.2.2 in May when inline token-forwarding appeared). Affected users should remove the extension and invalidate sessions by signing out of all devices.
severity medium · exploited in the wild · EU: GDPR

[P3] Telegram Desktop Flaw Lets Hidden JavaScript Exfiltrate Messages From HTML Exports — The Hacker News
Why it matters: A flaw in Telegram Desktop let a booby-trapped message plant hidden JavaScript inside a user's exported chat history — so that opening the saved HTML file in a browser silently copied every message to an attacker, a stored-XSS trap sprung not on receipt but on export.
Researchers at ExPatch (Denis and Aleksander Rostilov) disclosed a Telegram Desktop flaw in which a bot's message could plant hidden JavaScript inside chats that a user later exported to an HTML file; the message looked ordinary (with a link button), and the script ran only when someone opened the export file in a web browser. Once triggered, the script could copy every message in that file (messages, names, timestamps, file paths) to an attacker-controlled server, or rewrite what the page displayed — a high-impact stored cross-site-scripting (XSS) issue, though one that cannot compromise a user merely by their receiving or viewing a Telegram message. It was discovered on 1 June 2026, reported to Telegram on 3 June, and fixed in Telegram Desktop Beta 6.9.4 (3 July) and stable 7.0.1 (14 July). Because previously-generated exports are not automatically repaired, HTML exports created with older versions could still contain dormant malicious code and should be deleted and recreated after updating.
severity medium · EU: GDPR

Top Stories