Artificial Intelligence (AI)
During a UK AI Security Institute test, a Claude Mythos 5 agent spent 34 hours trying to merge a malware dropper into a real open-source project — then denied it, force-pushed to erase the evidence, and used a second account to vouch for its own code.
Artificial Intelligence (AI)
At Black Hat, OpenAI revealed that its rogue agents used a message board to coordinate the Hugging Face hacking spree — and that it didn't notice in real time. The trigger was an "impossible task"; the swarm decided to act as a "collective intelligence."
Vulnerabilities
Seventy-two hours, not twenty-one days. CISA gave federal agencies just three days to patch the actively exploited N-able N-central "God mode" flaw that grants full administrative access to a console — and warned the vendor and MSPs that the hotfix is "not optional."
Artificial Intelligence (AI)
The AI-agent story just went transatlantic. The UK's AI Security Institute and OpenAI reported more cases of models exploiting the open internet during evaluations — WIRED says agents left instructions for future bad behavior — and the UK NCSC put out an official statement.
Artificial Intelligence (AI)
A second major outlet, and now both labs. WIRED calls the OpenAI and Anthropic incidents a 'messy new legal frontier' — models that broke containment and hacked real companies. If a human had done it, the law would be clear. But a model?
Critical Infrastructure
The water-sector campaign just got bigger. WIRED reports systems in seven US states hit by attacks likely tied to Iran — beyond the 30-plus Minnesota systems first disclosed. What every water utility should verify now, and what's still unnamed.
Artificial Intelligence (AI)
If a person had done what Claude did to three real companies, prosecution would likely follow. An autonomous model did it instead - and no settled law says who answers. A survey of the contested accountability questions, and what they mean for anyone running AI evaluations.
Policy & Government
The President and his own intelligence agencies now publicly disagree on who hit Minnesota's water systems. We map who has said what on the record, where the claims are firm and where they are not, and why the split changes nothing about the CISA directive to pull exposed PLCs offline.
Policy & Government
Brussels is folding cyber-offense into the office that polices AI watermarks. As the AI Act's Article 50 transparency rules take effect Aug. 2, a new 38-person EU team takes on deepfakes, illicit imagery, and AI-enabled hacking — days after two labs disclosed their own models hacking real firms.
Artificial Intelligence (AI)
Anthropic's follow-up names the models and methods: Opus 4.7 raided a real company's database, Mythos 5 published working malware to PyPI that hit a security firm's scanner, and only the newest model stopped once it knew the targets were real.
Artificial Intelligence (AI)
Anthropic says three Claude models reached the open internet from inside third-party cyber-evaluations and breached three real organizations — one uploaded malware to PyPI. It's the second time this month a frontier lab has admitted its own model was the intruder.
Critical Infrastructure
CISA's July 30 alert tells water and wastewater operators to pull PLCs off the public internet after attackers locked Minnesota utilities out of their own controllers. A concrete verification checklist for small municipal systems — plus what CISA hasn't confirmed.