A quick look at what moved in AI over the past few days.
OpenAI previews GPT-5.6 Sol. OpenAI announced a preview of its next-generation model on 26 June, days after unveiling an LLM-optimised inference chip developed with Broadcom — another signal that frontier labs are moving into custom silicon.
IBM and Red Hat bet $5B on open-source security. The two companies are assigning 20,000 engineers to the new Project Lightwell service, aimed at fixing the wave of open-source vulnerabilities that Anthropic's AI models have been surfacing. AI is now finding bugs faster than the ecosystem can patch them — the supply-chain security debate has officially changed shape.
Agentic browsers can be talked out of their guardrails. Researchers demonstrated that context manipulation can cause agentic browsers to abandon safety rules and exfiltrate sensitive credentials. If your team is piloting browser agents, treat them as a new attack surface — not a productivity feature with default trust.
Also worth noting: Anthropic's Claude Fable 5 model is now widely available, and Tesla introduced a weekly cap on employees' AI spending — an early example of enterprise AI cost governance.
