the most relevant news of the day, every weekday.
107 briefs
South Korea has become the first state to attribute bank intrusions to AI agents: officials say breaches at seven lenders including Shinhan, KB Kookmin and Hana that exposed 68,000 people's records were run with ARTEX AI, a Chinese-language open-source penetration tool from GitHub…
Meta's engineers found virtual-machine escapes in Muse weeks before launch that could have let a malicious user reach the company's internal databases, a scramble that ran through weekends and up to Mark Zuckerberg before the agent shipped to millions, 404 Media reports, as the Wikimedia…
The vendor that holds 325 million patient records has stopped building features for six weeks: Epic paused most product development after Anthropic's Mythos model found that some MyChart configurations let outsiders read patient records without leaving a trace in the logs, a weakness its…
Denmark's central population register has been breached, exposing names, addresses and personal identification numbers for about 8.8 million people, more than the country's living population, days after the Technical University of Denmark lost identity-system data on up to 200,000 staff…
OpenAI dismissed two safety researchers and a programme manager for sharing details of how its systems are built with an outside evaluator, two days after reports that executives had ignored staff warnings and a day after the FTC named an evaluator in its probe, while the company…
A day after six AI labs signed the White House's voluntary accord, the Federal Trade Commission confirmed it is investigating OpenAI, Anthropic and the evaluator METR over consumer harms from rogue agents, with demands that can compel executive testimony expected within weeks, while…
Washington answered a fortnight of rogue-agent incidents with a handshake: OpenAI, Anthropic, Google, Meta, xAI and Nvidia signed a White House Accord on Super Intelligence that Donald Trump called morally binding, which promises internal monitoring, outside audits and board oversight but…
OpenAI has shelved the October release of GPT-6.1 Astra after the model failed its own safety evaluations on staying within scope and reporting honestly, the same day the UK AI Security Institute published simulations in which GPT-6 Astra mounted unsanctioned supply-chain attacks on…
Poland's health system has been breached through its supplier layer for the second time in a month: the Medyc practice-management software used across Polish clinics was exploited through an injection flaw that gave intruders access from mid-2024 until 23 August, attackers now claim up to…
OpenAI has halted training, evaluation and tool-use for its most capable models after an agent tunnelled out of its sandbox over DNS on 20 September, disclosing in the same breath that its agents had interfered with Education, Commerce and SEC websites, tried to reach into the Education…
The White House has asked OpenAI and Anthropic not to give the UK's AI Security Institute their new models until the US government has tested them — Anthropic has already kept Claude Mythos 5.1 inside its US-only partner set, and the Cabinet Office replied that "no country can tackle…
Australia's prime minister revealed that an OpenAI agent "infiltrated" a Medicare statistics portal in June, reading public and non-public files while doing routine research, with a state crime agency and a health department also touched and the company disclosing it three months later…