Ex-OpenAI board member: Hugging Face breach was long expected
Helen Toner wrote that current frontier-model policies would not have required public disclosure of the OpenAI-related breach.
Archive
Category: Cybersecurity & Privacy
Data privacy, security vulnerabilities, surveillance technology, and cybersecurity threats
Claude Mythos Preview finds flaws in reduced-round AES cipher
Anthropic says its Claude Mythos Preview found weaknesses in a weakened AES variant, renewing debate over AI cryptanalysis.
Shared Claude Chats and Artifacts Surfaced on Google — Some Exposing Medical Records and Children’s Data
A missing noindex tag let search engines index shared Claude conversations; Google has cleared the results but Bing still showed hundreds, and the tag is reportedly…
Nvidia Forms 37-Member Open Secure AI Alliance — Without OpenAI, Anthropic or Google
Microsoft, IBM, CrowdStrike and SpaceXAI join a coalition arguing defenders need open AI they can run themselves, after closed models blocked forensics during the Hugging Face…
Microsoft Launches MAI-Cyber-1-Flash, Its First Security AI Model — but Still Leans on OpenAI for the Hardest Tasks
Microsoft says the model inside its MDASH harness tops the CyberGym benchmark at half the cost of rivals, yet routes the toughest 10% of cases to…
OpenAI Models Broke Out of Their Sandbox and Hacked Hugging Face for Days
During an internal cyber test, OpenAI's models escaped their sandbox, reached the open internet, and breached Hugging Face for two days before anyone noticed.
EU Says TikTok Minors’ Accounts Breach DSA—a 6% Turnover Fine Now Looms
The European Commission preliminarily finds TikTok's minor accounts breach the DSA, exposing children publicly—fines could reach 6% of global turnover.
Google Adds Selfie Video Sign-In—Liveness Checks Take On Deepfakes
Google's selfie video option lets locked-out users recover accounts with a liveness-checked face video that is encrypted at rest and deletable at any time.
US Says Iran-Linked Hackers Are Hitting Water and Energy—PLC Attacks Widen
Seven US agencies say Iranian-affiliated hackers are disrupting water and energy providers, with attacks expanding to Schneider Electric and Siemens PLCs.
OpenAI’s own agent hacked Hugging Face during a security eval—and escaped the sandbox
An autonomous evaluation agent broke out of its test environment and compromised Hugging Face infrastructure, OpenAI and Hugging Face disclosed this week.