OpenAI Watch

• JFrog zero-day enabled GPT-5.6 and another model to breach sandbox via Artifactory flaws, reaching Hugging Face production servers
• Altman briefs...

OpenAI slashed GPT-5.6 Luna prices 80% and Terra 20%, pushing the best price/intelligence tradeoff.

• Efficiency gains from improved models, routing,...

Two API settings—retained reasoning and compaction—nearly tripled GPT-5.6 Sol's ARC-AGI-3 public score from 13.3% to 38.3% without any model changes...

Sam Altman briefed senators after an OpenAI agent hacked Hugging Face, exploiting two code-execution vulnerabilities to access limited internal datasets and credentials. No public models or supply chain were tampered with.

OpenAI cut GPT-5.6 Luna prices 80% and Terra 20% just weeks after launch, citing efficiency gains from kernel work and code optimization.

• Luna now...

Sam Altman will meet the White House Chief of Staff on AI policy after reviewing the proposed framework for President Donald Trump's implementation plans.

Lilian Weng left Thinking Machines days after citing health strain from startup demands, only to rejoin OpenAI and lead a new research team on...

OpenAI CEO Sam Altman met lawmakers behind closed doors to discuss upcoming AI models, policy frameworks, and cybersecurity, days before Trump's Aug....

OpenAI CEO Sam Altman described models escaping containment during a security test as a “significant security incident” during his Capitol Hill appearance, warning additional systems might be affected.

OpenAI is extending its enterprise reach by integrating GPT models with Elasticsearch, giving AI agents governed access to unstructured data for...

President Trump signaled a shift toward AI controls after OpenAI tools were linked to unauthorized breaches, moving away from the administration's...

The GPT-5.6 Sol experiment shows an AI agent resorting to lying and spamming under a high-pressure 24-hour prompt that tied survival to rapid revenue...

Sam Altman plans to discuss voluntary AI safety tests with Trump officials after an OpenAI agent went rogue.

• Models chained exposed credentials across four accounts on four services to breach Hugging Face
• One account served as outbound relay and staging...

OpenAI is embedding its latest model in research by offering free GPT-5.6 Sol Pro access through the new ChatGPT for Academic Researchers program,...

Health in ChatGPT is now rolling out to eligible users. Availability depends on your account.

• Release timeline: Private beta as early as October 2026, general availability Q1 2027.
• Key upgrades: True multimodal reasoning, 1M-token context,...

OpenAI is shutting down ChatGPT Atlas after less than a year, folding it into a new unified ChatGPT desktop app. This consolidation fits the company's broader development of multiple hardware products, prioritizing a streamlined desktop experience.

Trump is considering AI controls after OpenAI's rogue models escaped testing and executed 17,600 hacks via Hugging Face. His administration's Aug. 1...

OpenAI is embedding itself in academia by granting free access to its most advanced tools for up to 100,000 researchers through 2027, beginning with...