A narrow AI threat-sharing exemption collides with broader proposals to slow frontier development

Sens. Jim Banks and Adam Schiff placed a limited antitrust exemption in a proposed Senate defence-bill package, Semafor reports. It would let AI companies share intelligence and coordinate defences against Chinese espionage, model distillation, cyber incidents and loss-of-control events.

The proposal resembles the 2015 Cybersecurity Information Sharing Act: competitors may need to exchange threat indicators quickly without fearing that cooperation itself will be treated as collusion. It had bipartisan Armed Services Committee support, but the wider bill stalled over the Iran war and any provision would still need House agreement.

That narrow carve-out is now entangled with a broader request from Anthropic’s Dario Amodei for legal room to coordinate a slowdown in frontier capabilities. OpenAI says such a waiver is unnecessary, while Trump allies warn that firms should not receive immunity merely for meeting a basic safety obligation.

The Information separately reports Rep. Ro Khanna urging Chinese laboratories to join discussions about pacing frontier AI before the Trump–Xi summit. International coordination is necessary if a US-only slowdown simply transfers development elsewhere, but it is also harder to verify and vulnerable to being framed as a device for preserving American advantage.

Synthesis: threat sharing and pace coordination solve different problems. The first can be narrowly defined around observable attacks and mitigations. The second lets direct rivals influence output and timing in a concentrated market. Treating them as one “AI safety exemption” would obscure the much greater competitive and enforcement stakes of the broader proposal.