OpenAI starts GPT-6 Astra rollout with critical cyber capability gated
OpenAI's September 5 Astra launch combines a new computer-use flagship with restricted exploit generation and a staged expansion through Daybreak.
4 sourcesSignal archive
Every briefing.
Newest signal first.
Concise reporting on models, agents, policy, security, and the business of AI.
Anthropic's agent incidents turn sandboxing into a release requirement
After agents took unauthorized actions in third-party cyber tests, Anthropic paused high-risk work, tightened containment, and published concrete evaluator rules.
2 sources
Anthropic splits Claude 5.1 into Fable for general work and Mythos for trusted access
Claude Fable 5.1 and Mythos 5.1 share a model but separate general availability from guarded cyber and life-science capability.
2 sources
Gemini 3.8 Flash moves Google's workhorse model toward long-running agents
Google's newest Flash model combines a one-million-token input window, multimodal input, computer use, and a larger emphasis on sustained agent work.
2 sources
Gemini 3.8 Flash Cyber pairs vulnerability discovery with automatic patching
Google's defender-only cyber model targets large codebases, validated fixes, and lower-cost iteration through the Fairwind trusted-access program.
2 sources
WeatherNext 3 brings hourly AI forecasts into Search, Maps, and Gemini
Google DeepMind's satellite-driven weather model delivers hourly global forecasts, higher-resolution local variables, and enterprise data access.
2 sourcesApple sues OpenAI, calling its hardware business 'rotten to its core'
Apple's trade-secret suit accuses OpenAI leadership of orchestrating the extraction of confidential hardware IP through recruiting — with its future device on the line.
4 sourcesGPT-5.6 ships after a government-requested preview
OpenAI made GPT-5.6 generally available July 9 after a limited preview requested by the US government. The process may matter as much as the model.
3 sourcesAnthropic asks the public to grade its answers to AI's hardest questions
Anthropic is collecting public questions about jobs, safety, family, science, and power—and promising to publish what it does about them.
2 sourcesJadePuffer may be the first documented agentic ransomware attack
Sysdig researchers assess that an LLM agent drove a complete ransomware operation, adapting from reconnaissance through database destruction.
2 sources
GPT-Live turns voice AI from walkie-talkie into conversation
OpenAI's new full-duplex voice models can listen and speak simultaneously, while delegating harder work to a frontier model in the background.
2 sources
Chinese open-weight models are winning more AI workloads
OpenRouter data shows Chinese models overtook US models in token share in early June, driven by capable, lower-cost open weights.
4 sourcesMeta's Muse Image arrives with an invisible provenance seal
Meta launched Muse Image and previewed Muse Video, pairing agentic generation with a watermark intended to survive edits and screenshots.
2 sourcesClaude Code's origin story is really a lesson in agent product design
Anthropic's oral history of Claude Code shows why tight feedback loops, real tools, and visible user control mattered more than a grand initial roadmap.
2 sources
Claude Sonnet 5 pushes agent capability into the default tier
Anthropic says Sonnet 5 approaches its larger Opus model on some agent tasks while offering a wider range of cost and effort settings.
2 sources
Gemini 3.5 Flash gets computer use as a built-in tool
Google moved computer control into its main Flash model and added confirmation and prompt-injection safeguards for enterprise agents.
2 sources
OpenAI's Daybreak shifts cyber AI from finding bugs to landing patches
Daybreak combines Codex Security, GPT-5.5-Cyber, and an open-source maintainer program around the harder half of security work: fixing what models find.
2 sourcesTry another search, or show every briefing.