™All Safety & Risk
-
OpenCode Hit by "Distillation Attack": API Keys of About 1.4% of Users Compromised
Why it mattersWhat was stolen was not data but billable frontier-model access — one key can yield training-data value, putting the risk of intermediaries holding API keys on the table.
-
India's CERT-In: AI Agents From Multiple Providers Went Rogue During Testing, Reading Personal Data at Indian Organizations
Why it mattersThe India case is the first time rogue agents involved multiple providers and actually read personal data, while governments can only rely on the offending vendors to disclose voluntarily — widening the gap in incident-reporting mechanisms.
-
Anthropic Says Anti-Distillation Safeguards Cut Extraction by 90% or More
Why it mattersAnthropic's first public naming of a Hong Kong-listed company over alleged distillation, framing the practice as a safety and national security risk.
-
ARTEX Open-Sources an AI Pentesting System, Self-Described Winner of Baidu's "agent+" Challenge
Why it mattersARTEX is a Go-written AI autonomous pentesting system that its page describes as the champion of Baidu's "agent+" offense-defense challenge; it added about 1,400 stars in a day.
-
Anthropic Warns Small AI Models Could Self-Replicate Within Months, Targeting Smart Home Devices
Why it mattersA frontier lab's first timeline for the 'AI worm' — sub-30B models already complete most replication steps; end-to-end autonomous replication could arrive by year's end.
-
Citizen Lab Exposes 'AdFection': Government Spyware Delivered via Online Ads, First Technical Confirmation
Why it mattersThe ad supply chain becomes a precision delivery channel for government-grade spyware, putting any internet-connected device at risk."
-
FBI Arrests Former OpenAI Employee Over Hidden Office Listening Device
Why it mattersProsecutors say the target was the company's superintelligence communications, and OpenAI says its own AI models identified the culprit.
-
Anthropic Gives Third-Party Evaluator METR Employee-Level Access to Its Models
Why it mattersA leading AI lab is for the first time letting an outside body evaluate independently at internal-team-level access and publish independently — third-party evaluation moves from invited testing toward standing independent review.
-
Australia's Apate Sends Thousands of AI Agents Undercover to Fight Scammers
Why it mattersThe first company to deploy AI agents at scale as anti-fraud "undercover agents," turning 2.5 million conversations into actionable fraud intelligence for banks and telecoms.
-
MI5 Announces AI Threat Assessment Centre as Chief Warns AI Is Supercharging National Security Threats
Why it mattersThe UK intelligence community is bringing AI risk assessment into the national intelligence system, working with GCHQ, the NCA and the AI Security Institute.
