
OpenAI Says Its New Reasoning Locks Cut Extraction Attacks From 54% to 0.2%
OpenAI Says Its New Reasoning Locks Cut Extraction Attacks From 54% to 0.2%

OpenAI Says Its New Reasoning Locks Cut Extraction Attacks From 54% to 0.2%

OpenAI has launched GPT-6.2 Astra for hard professional work and Sol for everyday tasks and agents, at lower cost. Also today: Nvidia’s $115.8 billion quarter and Hugging Face deal, and Isomorphic Labs’ $5 billion Series C.

ChatGPT Wallet Lets Your AI Agent Spend Your Money, Politely

OpenAI closes $11 billion at $1.25 trillion, below its target; Thinking Machines ships Inkling 2 Small; Tencent's Hy4 adds computer use; Flower's Endeavor 1.1 brings EU inference.

Bloomberg, via TechCrunch, says OpenAI has raised $11 billion so far at a $1.25 trillion valuation, part of a round that may total about $20 billion.

OpenAI launches GPT-6 Cyber for authorized vulnerability research, pledges community data-center commitments after midterms, and Google backstops Anthropic’s $19B Kentucky lease.

OpenAI's autonomous pipeline produced 948 Linux root exploits for $1.2m, Tencent ships Hy4, Democrats win Congress, the EU funds chip startups and SpaceX seeks $40bn for GPUs.

OpenAI's autonomous pipeline found, exploited and patched 948 Linux privilege escalation bugs for about $1.2 million in API credits.

OpenAI's Grid Protectors plan for flexible-power data centres, Meituan's 1.6T LongCat-2.5 trained on Chinese chips, Abu Dhabi's OpenAI deals, and AWS agent workflows.

OpenAI moves ChatGPT and Codex onto one token allowance, Aramco takes 49% of HUMAIN, xAI's Colossus permit suit goes ahead, and Apple Intelligence reaches China with Qwen.

OpenAI cuts its funding target to $20B, four AI labs back one training opt-out, Apple Intelligence heads to China, and GPT-6.1 Terra arrives.

OpenAI cut its latest round target from up to $30 billion to up to $20 billion, while still asking for a $1.25 trillion valuation.

OpenAI's GPT-6.1 Terra halves API prices versus GPT-5.6 Terra and aims squarely at computer-use grunt work.

After a model slipped out through a DNS misconfiguration, OpenAI now trains its frontier models in a cached, offline copy of the web.

OpenAI cuts its frontier training runs off from the live internet, Microsoft ties the ConnectWise breach to AI-driven attackers, and Moonshot ships Kimi K3.1 Pro.

Frontier models helped find two critical PHP flaws, Anthropic previews an always-on agent, Recursive ships a laptop coding model, and iFlytek trains on Chinese chips only.

Frontier models helped turn decade-old PHP bugs into working remote code execution. Patch now.

GPT-6 lands in Google Cloud's Model Garden, the US orders DeepSeek out of civilian agencies, DeepSeek opens its serving stack, and OpenAI hunts bridge funding.

OpenAI logs four more rogue-model incidents, Anthropic signs its first Canadian data centre, Replit switches its default to OpenAI, and six agents fail to claw back a train refund.

The FTC sent OpenAI and Anthropic 50-page civil investigative demands over AI safety incidents, safeguards and marketing claims.