The model race is now a deployment-and-control race
OpenAI has opened a limited preview of GPT-5.6 Sol, Terra and Luna, with Sol reserved initially for a small group of trusted partners after discussions with the US government. The move combines a major capability jump—new agentic coding, biology and cybersecurity evaluations—with a release process that makes government access a condition of first availability. That matters because developers and defenders get a stronger system, but not yet on normal terms. OpenAI says broader availability is planned in the coming weeks, while insisting that government pre-clearance should not become the permanent model-release default. Read OpenAI’s preview announcement.
What else matters
GitHub Spark enters its final month
GitHub stopped accepting new Spark users and new app creation on 4 August. Existing users have until 31 August to export their apps; deployed apps will continue running, but Spark’s llm() function already stopped working after GitHub Models retired on 30 July. Teams need to export code and replace inference themselves. Read GitHub’s deprecation notice.
Anthropic adds a global-affairs chief as policy pressure rises
Anthropic’s newsroom lists Mariano-Florentino (Tino) Cuéllar as its incoming Chief Global Affairs Officer. The appointment lands as frontier labs face more scrutiny over model releases, safety commitments and government relationships. It is a personnel change, not a new policy, but it signals that external affairs is becoming core operating infrastructure for model companies. See Anthropic’s newsroom announcement.
The EU’s AI Act transparency clock is running
The European Commission says the AI Act’s transparency rules come into effect in August 2026. Providers and deployers should be preparing for obligations around informing users when they interact with AI and marking synthetic content, alongside the Act’s broader risk-based duties. The practical implication is a compliance deadline that reaches product, documentation and content pipelines—not just legal teams. Check the European Commission’s AI Act guidance.
OpenAI’s safety claim comes with a phased-access caveat
GPT-5.6 Sol is described as OpenAI’s strongest model yet, with stronger safeguards for sensitive cyber requests and repeated misuse. OpenAI also says benchmark thresholds cannot capture every combination of model and tool use, and is withholding broad access while it tests the system. The near-term question is whether the extra review improves real-world assurance without turning controlled access into a durable gate. Read the model preview and safety details.
