Change of course on AI: Donald Trump is thinking about regulating artificial intelligence in the United States

a government oversight process White House officials would have discussed it the previous week with leaders from Anthropic, Google, and OpenAI, according to Bloomberg.

Claude Mythos, the model that accelerated the reflection

The trigger has a name. On April 7, Anthropic introduced its Claude Mythos Preview model, along with the Glasswing project. This security testing consortium brings together a handful of carefully selected organizations. The model has not been made public, and for good reason.

In seven weeks of testing, Mythos identified more than 2,000 previously unknown software vulnerabilities across all major operating systems and browsers. It rediscovered a bug that had been around for 27 years in OpenBSD. Its rate of reproducing functional exploits on the first attempt reaches 83%. The British AI Safety Institute (one of the only non-American organizations to have access to the model) confirmed that the model represented a leap forward compared to previous generations in terms of offensive capabilities in cybersecurity.

Also read: « The End of Security Flaws » – Mozilla fixed 271 vulnerabilities in Firefox thanks to Claude Mythos

According to Reuters, these performances would have directly fueled the discussions at the White House. The spokesperson responded succinctly that « any announcement would come directly from the president » and called the rumors « speculation ».

Washington is moving (unwillingly) towards the European model

On August 2, 2026, the European Union will apply the full obligations of the AI Act to high-risk AI systems. Biometrics, recruitment, health, financial services, critical infrastructure: all sensitive uses will have to go through a documented compliance assessment. Fines can reach 35 million euros or 7% of global turnover.

In France, the CNIL has been designated as the national authority for coordinating the implementation of the text. The government has also invested 400 million euros in nine AI clusters to support the industry. The obligation to train on AI issues (Article 4 of the regulation) applies to all companies since February 2025.

On the American side, the lack of a federal framework has led to a patchwork of laws that the states have not yet fully regulated. 38 states have adopted their own AI laws between 2024 and 2026. Colorado requires annual audits on algorithmic discrimination. California has passed a successor to SB 1047 (itself a transposition of the Biden administration's law repealed by Trump upon his arrival at the White House). Texas and Florida have taken the opposite path, with laws protecting developers from any state constraints.

For European companies operating in the United States (and vice versa), the prospect of a dual compliance regime is becoming clearer. The AI Act imposes a centralized framework. Washington oscillates between deregulation and case-by-case control. Navigating between the two promises to be as complex as it is costly.

👉🏻 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source : The New York Times