Today AI News Roundup: GPT-5.6 Sol Ultra, US Deregulates Anthropic, and Geopolitical Tensions Escalate

OpenAI launches GPT-5.6 Sol Ultra in Codex

OpenAI released GPT-5.6 Sol Ultra through the Codex CLI tool. It costs about USD 0.15 per million input tokens on a pay-as-you-go basis. The model is a coding-focused variant of the Sol line. It scored well on SWE-bench and SWE-agent, two benchmarks that measure how well AI handles real-world software engineering work. In one demo, the model found a GitHub issue, wrote a fix, and deployed it to a cloud environment. No human intervened at any step.

US lifts restrictions on Anthropic's Fable and Mythos

The US government dropped export controls and safety deployment restrictions on Anthropic's Fable 5 and Mythos models. Companies can now deploy and distribute them globally. A federal review found that the models carry no more risk than other commercial AI systems. The change helps Anthropic's enterprise customers outside the US. Industries that rely on Anthropic's safety features stand to benefit the most.

Alibaba bans Claude Code citing security concerns

Alibaba told its employees to stop using Anthropic's Claude Code tool. An internal memo cited alleged backdoor security risks and told staff to remove any Claude Code-related software from work devices. The ban comes as US-China technology tensions rise. Other Chinese companies have taken similar steps to restrict AI development tools. The global AI market is splitting along geopolitical lines.

Why this matters

These stories point in different directions. Frontier models are getting cheaper and easier to access. At the same time, geopolitical tensions are creating new barriers. For companies that build on AI, the message is straightforward. Where you operate will determine what models and tools you can use. And the regulatory environment can shift without warning.