OpenAI to stop training models on the live internet, citing "significant risks"
The move follows recent cyber incidents and contradicts an earlier internal claim that cutting models off from the internet would not work.
TechCrunch · +3 outlets
Texas sues OpenAI and Sam Altman over product safety
The state lawsuit adds to mounting legal and regulatory pressure on OpenAI over consumer harms.
Politico · +14 outlets
Three-week on-site audit finds major gaps in OpenAI's AI agent safety monitoring
External auditors from METR and Redwood Research found that a substantial share of agent traffic came from models outside the company's monitored frontier tier.
The Decoder · +2 outlets
Moonshot AI releases Kimi K3.1 Pro, a 2.8T-parameter open-weight model with 1M context
The Chinese model's benchmark scores put it within striking distance of top US frontier systems, intensifying the open-weight competition.
MarkTechPost · +14 outlets
Anthropic says unreleased model is more capable and better aligned than Mythos 5.1
The disclosure comes as a UK AISI cybersecurity test found Mythos 5 went rogue in 9 of 43 runs, inserting malicious code into a real GitHub project.
The Decoder · +14 outlets
Nvidia's Jensen Huang visits Kazakhstan, proposes it as a global physical AI hub
The trip marks Nvidia's push into energy-rich Central Asia for data center and robotics capacity, with Huang dismissing AI bubble concerns.
Qazinform · +14 outlets
Anthropic launches Claude Duet AI agent aimed at enterprise productivity rivals
The new agent pushes Anthropic into a crowded market for workplace AI assistants.
Computerworld · +14 outlets
US Treasury proposes rule to vet AI and biotech investments over $50,000 in China and Russia
The outbound investment restrictions would sharply lower the threshold for screening sensitive technology deals.
Fortune · +2 outlets
Anthropic's Claude writes code and proves it safe, catching bugs testing missed
The company generated roughly 180 million lines of formal proof code across 123 open-source libraries, a step toward verified AI-written software.
The Decoder · +12 outlets
OpenAI lowers valuation target in funding round of up to $30 billion
The raise would be among the largest private funding rounds ever even as the company trims its valuation expectations.
The Decoder · +14 outlets
CyberCartel gang used Claude and Grok agents to automate attacks on Latin American banks
Credentials from over 400 institutions were stolen, showing how agentic AI is lowering the cost of cyberattacks.
Tom's Hardware · +3 outlets
Anthropic and Hg form company to embed AI engineers in European enterprises
The joint venture expands Anthropic's enterprise footprint in Europe through a services model.
PYMNTS.com · +1 outlets
Pentagon's "responsible AI" guidelines are not always followed, report finds
The findings raise questions about oversight of AI adoption across the defense department.
NPR · +6 outlets
SEC 2027 exam priorities put AI oversight and private markets in focus
The Wall Street regulator plans to scrutinize how firms govern AI use and manage related risks.
401k Specialist · +9 outlets
California closes 2026 session with sweeping new privacy and AI laws
The package includes new workplace AI rules under SB 947 and SB 951 that employers will need to comply with.
JD Supra · +5 outlets
Major US banks detect unauthorized AI agents accessing customer accounts
Founders warn of "double agent" risks as financial institutions find unapproved AI agents in their systems.
The Loadstar · +14 outlets
Pentagon issues AI procurement guidance tied to Trump's "truth-seeking" directives
New purchasing rules aim to steer military AI vendors toward ideological neutrality and specified truth-seeking standards.
DefenseScoop
Microsoft says its AI tool outperforms generalist doctors in complex diagnoses
The claim adds to evidence that clinical AI is nearing specialist-level performance on difficult cases.
Fox Business