Fired OpenAI Trio: Don't Let Models Outwit the Monitor

Tomek Korbak, Mikita Balesni and Jasmine Wang, fired last week over handling sensitive information, tell OpenAI's board the industry cannot safely ship models it cannot monitor - and OpenAI's own GPT-6 Astra card concedes the class could evade its chain-of-thought monitors under adversarial conditions.

The RaceCompanies, models and money

GPT-6 Swaps the Answer for an Interface

GPT-6 reaches all 1.2 billion weekly ChatGPT users with Intelligent UI - charts, forms, tappable buttons and generated mini-apps instead of text - and it answers while still thinking, cutting waits 44 percent.

The LabResearch, benchmarks and safety

Drake Calls 'Bunker Mode.' Buterin Says Don't Panic.

Ethereum Foundation researcher Justin Drake wants a controlled migration of funds to fresh addresses, saying AI-accelerated math could break ECDSA in months not years - Buterin backs taking it seriously but tells holders not to scramble today.

The WorkbenchTools and agents

Cursor Puts Your Local Agents on Your Phone

Antigravity Adds Agent Automations

Copilot Sandboxing Hits GA

The FightLaw, policy and power

Korea Bank Hacker Left His Claude Logs Open

AI Helped Write the AI Hacking Apology

AI Will Sort Pentagon Secrets

The FalloutJobs, money, daily life

Firmus Cuts Its $44BN Price. May Pull the Float.

Sources tell Guardian Australia the AI datacentre company is slashing its near-$44bn valuation to find buyers - or may shelve what was billed as Australia's biggest listing in decades - and it withdrew from a parliamentary AI inquiry the same day.

Co-op's AI Scores Every Call to the Bereaved

ChatterOpinion, drama and noise

Musk Puts Claude and Suno Inside Grok Bot