US presses Meta on AI security reviews amid copyright fight

According to a New York Times report relayed by Reuters, the US government is pressing Meta to agree to AI safety reviews as national-security concerns around frontier models rise. The pressure reflects a broader regulatory turn — the same environment in which the US earlier ordered Anthropic to restrict Fable 5 — and signals that Washington increasingly wants pre-deployment scrutiny of powerful models, including open-weight releases where control is harder once weights are public.

On the legal front, Meta is fighting a bid by 13 authors to quickly appeal a ruling that its use of copyrighted works to train Llama constituted fair use. The fair-use ruling was a win for Meta, but an expedited appeal could put the question before a higher court sooner, with industry-wide implications for how training data is sourced. The outcome matters far beyond Meta given how much of the industry relies on similar fair-use arguments.

Commercially, Meta used Cannes Lions to unveil a creative tool that learns a brand's identity and generates ads on the fly — part of its effort to justify a roughly $145B capex bill by tying AI investment to advertising revenue, its core business. Separately, Meta continues advancing open agentic infrastructure via the OpenEnv framework built with Hugging Face, and Yann LeCun used UN Open Source Week to champion 'AI sovereignty' and open foundation models.

The through-line is a company simultaneously courting regulators, defending its training practices in court, and racing to monetize — a precarious balance. Watch whether Meta accepts the safety-review framework, which would set a precedent rivals would be pressured to match.