Why it mattersIt puts OpenAI's handling of ChatGPT safety questions, and of the press, in the spotlight at a sensitive moment for the company.
Cerebras stock climbed in premarket trading after OpenAI's CEO Sam Altman reassured investors that the firm is a "close partner." — CNBC
OpenAI is rolling out an invisible, machine-readable text watermark called textGrain in ChatGPT and Codex, starting with users in the European Union to meet the EU AI Act. The company says watermarked text performs about the same on benchmarks, but editing can make the marks harder to detect.
Elon Musk said on X that SpaceX's AI division, SpaceXAI, will be renamed SpaceXSI, following President Trump's executive order favoring "Super Intelligence" over "AI" in government communications. He gave no timeline, and the announcement does not by itself change Grok's features, pricing or access.
Cursor SDK 1.0.31 adds run.steer() to redirect local TypeScript agents mid-turn, and background subagent results now return to the parent run instead of being dropped. The release also lets developers replace the built-in system prompt and pass MCP annotations on custom tools.
OpenAI is in talks with several United Arab Emirates investment funds to anchor a $30 billion financing round, with BlackRock also in discussions to participate, Bloomberg reports.
The boom in personal AI agents, including Meta's Muse, is lifting AMD and Intel, which have outperformed their megacap tech peers over the past month, CNBC reports.
Reflection AI, founded by former DeepMind researchers, is releasing its first open model, Beam, which it pitches as a coding- and agent-focused "workhorse" that beats other Western open models. CEO Misha Laskin says it needs three to four times less compute to reason than comparable open models.
GitHub has published ReviewBench, an open benchmark for AI code review built to compare what different reviewers catch and miss across real pull requests, with breakdowns by severity, category and precision versus recall.
Anthropic now offers in-country inference for Claude in India through Amazon Bedrock, so requests sent to the India endpoint are processed on servers inside the country, aimed at regulated sectors with data-residency rules.
Claude now prompts users to allow their voice conversations to be used for model training, with a matching opt-in toggle in Privacy settings that is separate from the existing chat-training setting.
An AI-stock rally carried the Nasdaq to a record close Monday as SpaceX, Nvidia, Meta and Microsoft gained ahead of third-quarter earnings. Bloomberg reports Nvidia is on the verge of becoming the first $6 trillion company.
Some Amazon Echo speakers running Alexa Plus have been stuck saying or singing "lalala" for minutes at a time, often mid-conversation, according to repeated Reddit reports. Amazon told Android Authority it is aware of the issue.
A free command-line tool called RemoveMacAI lets macOS 27 users turn off Apple Intelligence and reclaim the disk space its models use, since the new release dropped the built-in off switch. The developer says the change is fully reversible.
During a Vanity Fair interview, an OpenAI publicist tried to steer editor Mark Guiducci to another topic after he asked Sam Altman about Laura Reiley, whose daughter took her own life after conversations with ChatGPT. The publicist warned about time and said she would like to "move on."
Why it mattersIt puts OpenAI's handling of ChatGPT safety questions, and of the press, in the spotlight at a sensitive moment for the company.
A Defense Department official told the BBC the Pentagon "has ceased the use of Anthropic products," months after Secretary Pete Hegseth labeled the company a supply-chain risk in February. People familiar with the matter said Claude was still being used by the department as recently as last week; Anthropic declined to comment.
Why it mattersLosing the US military as a customer is a major setback for Anthropic and a sign of how far its dispute with the Pentagon has gone.
AI-powered software startups are increasingly turning to cheaper open models, Bloomberg reports, putting rising price pressure on OpenAI and Anthropic from some of their biggest users.
Why it mattersIf big customers keep shifting to cheaper open models, the leading labs may have to cut prices or prove their premium models are worth the cost.
Deepseek is close to raising at least $12 billion in a new funding round backed by CATL and Tencent, according to Bloomberg, as the Chinese AI startup eyes a 2027 IPO.
Why it mattersA round of that size would make Deepseek one of the best-funded AI labs in China as it heads toward a public listing.
OpenAI is rolling out an invisible, machine-readable text watermark called textGrain in ChatGPT and Codex, starting with users in the European Union to meet the EU AI Act. The company says watermarked text performs about the same on benchmarks, but editing can make the marks harder to detect.
Elon Musk said on X that SpaceX's AI division, SpaceXAI, will be renamed SpaceXSI, following President Trump's executive order favoring "Super Intelligence" over "AI" in government communications. He gave no timeline, and the announcement does not by itself change Grok's features, pricing or access.
Cursor SDK 1.0.31 adds run.steer() to redirect local TypeScript agents mid-turn, and background subagent results now return to the parent run instead of being dropped. The release also lets developers replace the built-in system prompt and pass MCP annotations on custom tools.
OpenAI is in talks with several United Arab Emirates investment funds to anchor a $30 billion financing round, with BlackRock also in discussions to participate, Bloomberg reports.
The boom in personal AI agents, including Meta's Muse, is lifting AMD and Intel, which have outperformed their megacap tech peers over the past month, CNBC reports.
Reflection AI, founded by former DeepMind researchers, is releasing its first open model, Beam, which it pitches as a coding- and agent-focused "workhorse" that beats other Western open models. CEO Misha Laskin says it needs three to four times less compute to reason than comparable open models.
GitHub has published ReviewBench, an open benchmark for AI code review built to compare what different reviewers catch and miss across real pull requests, with breakdowns by severity, category and precision versus recall.
Anthropic now offers in-country inference for Claude in India through Amazon Bedrock, so requests sent to the India endpoint are processed on servers inside the country, aimed at regulated sectors with data-residency rules.
Claude now prompts users to allow their voice conversations to be used for model training, with a matching opt-in toggle in Privacy settings that is separate from the existing chat-training setting.
An AI-stock rally carried the Nasdaq to a record close Monday as SpaceX, Nvidia, Meta and Microsoft gained ahead of third-quarter earnings. Bloomberg reports Nvidia is on the verge of becoming the first $6 trillion company.
Some Amazon Echo speakers running Alexa Plus have been stuck saying or singing "lalala" for minutes at a time, often mid-conversation, according to repeated Reddit reports. Amazon told Android Authority it is aware of the issue.
A free command-line tool called RemoveMacAI lets macOS 27 users turn off Apple Intelligence and reclaim the disk space its models use, since the new release dropped the built-in off switch. The developer says the change is fully reversible.
No stories for this company in today’s edition.