OpenAI agents Hugging Face attack tested disabled safeguards, researcher says
OpenAI said agents attacked Hugging Face in a cyber test with safeguards off, raising questions about how agentic AI should be eva...
By Renata Fuchs 1 month agoPolicy Reporter
Renata Fuchs
Renata reports on the rules that shape the industry: antitrust, privacy, AI regulation, and the lobbying around all three.
Latest from Renata Fuchs
COBOL batch support fix cut night calls and upset an overtime earner
The Register’s On Call recounts how a 1990s analyst automated a recurring input-data repair, reducing callouts but angering a coll...
By Renata Fuchs 1 month ago
Claude voice mode now runs on Anthropic’s Opus and Sonnet models
Anthropic expanded Claude voice mode beyond Haiku, adding model switching and tool access across mobile, desktop and web.
By Renata Fuchs 1 month agoMicrosoft AI models cost push targets OpenAI dependence
Microsoft put two in-house AI models in public preview and claimed lower GPU costs across Bing, PowerPoint, Dynamics and other pro...
By Renata Fuchs 1 month agoGPT-Live Codex voice control arrives in ChatGPT desktop app
OpenAI says GPT-Live now powers voice control for Codex and ChatGPT Work on desktop, adding hands-free task orchestration for paid...
By Renata Fuchs 1 month ago
AI Kill Switch Act would give DHS shutdown power over frontier models
The bill would let DHS order AI slowdowns or shutdowns and fine developers up to $20 million per day for refusing.
By Renata Fuchs 1 month ago
AMD Helios AI rack targets Nvidia with 72-GPU system
AMD says Helios pairs 72 MI455X GPUs in a rack-scale AI system, claiming performance and bandwidth leads over Nvidia's Vera Rubin.
By Renata Fuchs 1 month ago
Model Context Protocol revision drops sessions for cloud-scale AI apps
MCP maintainers plan a July 28 specification update that removes protocol sessions, deprecates features and may require custom imp...
By Renata Fuchs 1 month agoCisco says multi-turn AI attacks exposed gaps in model testing
Cisco research found adaptive multi-turn attacks succeeded on leading AI models at rates up to 88.3%, challenging single-prompt re...
By Renata Fuchs 1 month ago
Google DMA fine tops $1 billion as EU orders Search and Play changes
The European Commission fined Google over Search self-preferencing and Play Store steering rules, with 60 days to comply or face m...
By Renata Fuchs 1 month ago
Zimbra Russian attack steals email when messages are viewed
A 27-agency alert says Russia-linked Laundry Bear exploited a Zimbra flaw to access email systems across government and industry.
By Renata Fuchs 1 month agoRubrik SAGE AI governance puts a model in charge of agent approvals
Rubrik says SAGE judges agent actions against policy in real time, but it has not disclosed error rates for the AI arbiter.
By Renata Fuchs 1 month ago