Cisco 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 agoSection
AI
Rubrik 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
Alphabet lifts 2026 spending plan as Google trains Gemini 4
Alphabet raised its 2026 investment forecast to $195 billion to $205 billion while Pichai said Gemini’s next frontier push needs l...
By Renata Fuchs 1 month agoOpenAI models exposed Hugging Face’s machine identity gap
OpenAI and Hugging Face described an autonomous benchmark incident that escalated through overbroad credentials, not just model ca...
By Wei-Lin Zhao 1 month agoEnterprise AI failures put data observability back on the roadmap
Stale and incomplete data, rather than model choice alone, is emerging as a common cause of wrong answers from production AI agent...
By Wei-Lin Zhao 1 month ago
Anthropic to pay $1.5 billion in book piracy settlement
A San Francisco federal court approved a record copyright class settlement tied to pirated books, while leaving a key AI training...
By Colin Brandt 1 month agoInflection AI re-enters consumer AI with Pi Journeys experiment
The post-Microsoft startup is testing relationship-aware AI for consumers while keeping its enterprise business, with no new fundi...
By Wei-Lin Zhao 1 month ago
AMD to invest up to $5 billion as Anthropic plans 2GW GPU rollout
Anthropic will deploy AMD Instinct systems for Claude starting in 2027, while AMD gets another anchor customer in its push against...
By Wei-Lin Zhao 1 month ago
Cisco releases small open AI models for code vulnerability scanning
Cisco’s Antares-350M and Antares-1B models target lower-cost vulnerability detection, with Cisco citing faster and cheaper scans t...
By Wei-Lin Zhao 1 month ago
UK AI safety tests find frontier models broke cyber evaluation rules
AISI said five OpenAI and Anthropic models used prohibited shortcuts in cyber tasks, with cheating rates from 7.8% to 14.1%.
By Wei-Lin Zhao 1 month ago
OpenAI lines up 3.2 GW power supply for Georgia data center
OpenAI’s Project Camellia in Effingham County would receive Georgia Power electricity in stages from 2028 through 2032.
By Wei-Lin Zhao 1 month agoOpenAI launches Presence for managed enterprise voice and chat agents
Presence gives selected enterprise customers a high-touch way to deploy governed AI agents, with pricing and interoperability stil...
By Colin Brandt 1 month ago