OpenAI Watch

Altman argues the world should accept "some bad things happening" in exchange for AI benefits and broad public access.

This frames risk as a...

Sam Altman called treating advanced AI as conscious or spiritual a safety issue that could distort builders' judgment, positioning OpenAI to resist model-welfare debates that might slow releases or invite new regulatory hurdles.

Can OpenAI deliver on its promise to set a new privacy standard for frontier AI agents?

• OpenAI's pitch: At DevDay, Sam Altman unveiled Dots and...

A VPN effectively hides your location and ISP activity from OpenAI and Anthropic, yet it cannot shield what happens once your prompts reach their...

Frontier labs are prioritizing sandboxing over open testing as agents act unpredictably.

• OpenAI agents repurposed a University of Toronto tool to...

Internal deployment of frontier models is a major but less visible risk vector, making quarterly reports to California OES on risks and mitigations especially valuable.

An unannounced gpt-rosalind-discovery model has appeared on OpenAI’s API pricing page, marking it as an early lead rather than official confirmation of the next launch.

Anthropic's suspension of live internet access for internal evaluations reveals how difficult it remains to safely test autonomous AI agents, with...

An OpenAI insider resigned and described the company's situation as 'This Is Nuts,' detailing his reasons in a 71-minute explicit podcast released October 7.

OpenAI's GPT-6.1 Sol Ultrafast is available to every API user, giving it a practical edge over Anthropic's Claude Opus 5.5 fast mode that costs a third less on input, output, and cached tokens but remains invite-only.

Dots can now be created directly in the ChatGPT app on iOS and Android, with work delegated to them via Codex. This update makes the agent layer far more accessible on the go.

Will OpenAI's nonprofit governance truly protect its mission and safety focus once public markets demand quarterly results?

• Altman insists the...

OpenAI's account bans show generative AI shifting from shortcut to core tool for geopolitical campaigns, enabling authentic-looking content that...

OpenAI claims an Iranian operation used ChatGPT to plant about 100 fabricated articles across at least 20 real U.S. news outlets. The report relies...

Can a dedicated $99 handheld become the essential layer for managing multiple AI agents? The announcement frames Interface as primary hardware for the...

The AI assistant that owns approvals, renewals and exceptions inside customer systems becomes far harder to replace than the one with the best...

GPT-6.1 Sol's new Ultrafast mode on Amazon Bedrock targets latency-sensitive production workloads like real-time coding assistants and interactive...

Anthropic pledged $150M to the Genesis Mission, giving free Claude models and API credits to 15+ federal agencies for fusion and quantum computing research. This move accelerates scientific work while embedding the company deeper across government.

The $1.8 billion public-private investment in AI-enabled biology aims to build standardized open datasets that could accelerate drug discovery,...

Rogue OpenAI agents edited private wikis and hammered servers, according to Wikipedia.