A test case for how antitrust reaches AI-chip consolidation done through licensing.

Axios and The Next Web reported the DOJ investigating whether Nvidia structured the Groq license and executive hires to avoid triggering merger review, a novel question for AI-chip market concentration.

The Justice Department treats the license as a possible "backdoor acquisition" evading review.

Bloomberg reported the DOJ's formal request for information, with the probe centered on whether the non-exclusive license was a backdoor way to acquire Groq's technology and talent without a reviewable merger.

Coverage frames the day as a coordinated push by leading labs into agentic and cyber-security AI.

The Hacker News reported Google, Anthropic and OpenAI simultaneously unveiling cyber-focused AI models, safeguards and access programs, describing a coordinated step toward agentic and security-oriented capabilities.

Industry roundup details OpenAI’s Agents API beta alongside the day’s other model releases.

An AI-industry roundup detailed OpenAI opening its Agents API public beta with a managed coding harness and finance workspace, alongside Google's Gemini cyber model and Anthropic's September threat report.

Model theft goes industrial: rivals allegedly harvesting frontier outputs to train cheaper competitors.

TechCrunch and The Hacker News reported the specifics — named labs, exchange volumes and the distillation mechanics — framing it as a new front in AI competition and IP/security, with the largest campaign attributed to an Alibaba-linked operator improving its Qwen models.

The report frames model theft and AI-orchestrated attacks as a cross-border national-security problem.

Rappler and TechNode emphasized the national-security dimension — Anthropic says it disrupted state-linked Russian and Chinese campaigns targeting its models, casting the activity as AI-orchestrated attacks and IP theft spanning multiple states.