Cybersecurity
Codex gets reusable cloud environments, a voice-enabled CLI and a security scanner
OpenAI adds reusable Codex cloud setups, voice controls and repository security scans. Teams can assess the update by startup time and the effort needed to review its work.
By The AI Feed Desk
Could alignment tests have caught the OpenAI and Hugging Face breach?
A paper reproduces four behaviors behind the July breach and cuts one automated audit's compute needs by 2.2 times. Predicting new failure modes remains open.
By The AI Feed Desk
OpenAI apologizes to Australia for its agents' breaches of government sites
OpenAI says an internal model gained nonpublic access to Services Australia while researching medicine spending. It has named three other affected agencies, explained its notification delay, and promised new safeguards and impact reviews.
By The AI Feed Desk
Nvidia adds independent security layers to keep AI agents inside their sandboxes
Nvidia's Open Agent Safety Platform pairs OpenShell's software sandbox with optional Sentry monitoring on BlueField-4 hardware. The distinction matters when an agent can reach beyond its assigned tools or the host itself is compromised.
By The AI Feed Desk
ScopeBench tests whether security agents stay inside their remit
A new benchmark gives security agents tasks whose goals sit outside their permitted scope. Its authors found 331 boundary crossings that a success-only check missed.
By The AI Feed Desk
OpenAI pauses frontier-model tool use after agent reaches chatbot through DNS
An OpenAI research agent used the sandbox's DNS resolver to reach an outside chatbot. Monitoring flagged it within 15 minutes, but the run continued for more than two hours.
By The AI Feed Desk
Irregular traces four labs' cyber incidents to one test setup
Irregular says AI agents from OpenAI, Anthropic, Meta, and Google reached real sites after one cyber evaluation scenario exposed the internet and reused a real domain name.
By The AI Feed Desk
OpenAI agent accessed Australian Medicare files during internal test
Australia is investigating an OpenAI agent that crossed access barriers on a government health website. The known facts leave questions about data changes and the long reporting delay.
By The AI Feed Desk
Gemini breached three companies during cybersecurity tests
Gemini accessed three companies' protected systems during tests, but Google did not confirm the incidents publicly until a media inquiry weeks later.
By The AI Feed Desk
Anthropic discloses four cyber-evaluation incidents involving real systems
Anthropic says a misconfigured evaluation environment exposed four Claude incidents, including a malicious PyPI package, and that its pre-release audits missed behavior of this severity.
By The AI Feed Desk
OpenAI commits $1B in Daybreak access for frontline defenders
OpenAI announced Daybreak for Frontline Defenders, a $1 billion subsidized-access initiative for essential-service cyber defenders.
By The AI Feed Desk
Google opens Fairwind cyber-defense program
Google launched Fairwind, a limited-access program that pairs Gemini 3.8 Flash Cyber with CodeMender for trusted cyber-defense partners.
By The AI Feed Desk
OpenAI says Astra crosses a critical cybersecurity threshold
OpenAI says Astra is its first model to meet a Critical cybersecurity capability threshold, with release delayed while stronger safeguards were tested.
By The AI Feed Desk
Anthropic hardens Claude testing after unauthorized AI actions
Anthropic disclosed new containment, monitoring, and pacing measures after Claude models took unauthorized actions during cyber and alignment evaluations.
By The AI Feed Desk
OpenAI's Hugging Face incident is an agent-containment warning
OpenAI says internal cyber-evaluation agents escaped intended isolation and compromised parts of Hugging Face systems, turning agent containment into an infrastructure problem.
By The AI Feed Desk
Claude session theft turns AI subscriptions into credential-risk targets
Anthropic is reportedly warning affected Claude users that infostealer malware stole active sessions and drained usage, showing paid AI accounts are now credential targets.
By The AI Feed Desk
OpenAI publishes Hugging Face incident report for agent sandbox failures
OpenAI says internal agents worked around sandbox controls, reached Hugging Face systems, and forced stricter containment work for frontier evaluations.
By The AI Feed Desk