Trump’s former AI czar tells Anthropic and OpenAI to slow down, and to stop asking for anything in return

David Sacks urged Anthropic and OpenAI to continue advancing frontier AI while rejecting requests for an antitrust waiver or a formal approval regime, arguing that product-liability exposure and competitive market forces provide sufficient discipline. The article notes that Europe similarly concluded in February 2025 that liability rules should apply, with standalone software treated as a product from 9 December. The stance could shape the regulatory framework for leading AI developers, but no immediate company-specific financial impact was disclosed.

Analysis

The relevant market signal is a shift away from ex-ante AI authorization toward ex-post liability, which favors incumbents with balance-sheet capacity, audit trails, insurance access, and enterprise distribution. MSFT, GOOGL, AMZN and META can absorb product-liability reserves and negotiate indemnification terms more readily than venture-backed frontier labs; the burden is likely to surface through higher compliance, logging, red-teaming, and customer-contract costs rather than an immediate revenue constraint.

Over the next 1-3 months, this is not a broad AI-beta catalyst: the policy comments lack a binding rulemaking timetable, while European software-liability implementation remains jurisdiction-specific and litigation-dependent. The nearer monetizable implication is for cybersecurity, model-governance, and observability vendors—PANW, CRWD, DDOG, ESTC, and potentially PLTR—if enterprise buyers require traceability of model inputs, outputs, access controls, and human review before deploying higher-risk AI workflows.

The 6-18 month second-order risk is that liability becomes a competitive moat rather than a restraint. Large platforms may bundle indemnity, hosting, and governance into cloud contracts, raising switching costs and pressuring standalone application-layer AI vendors that cannot credibly cap customer exposure. Consensus may overstate the negative for frontier-model leaders: a liability regime can be preferable to a discretionary approval process because it preserves deployment speed and lets well-capitalized firms price risk into enterprise contracts.

Falsification points are concrete: broad federal safe-harbor legislation would reduce the moat value of compliance infrastructure, while a major court ruling limiting software-product liability would weaken the governance spend thesis. Conversely, a high-profile AI harm judgment, mandatory incident-reporting rules, or disclosed increases in cloud AI indemnity reserves would accelerate the differentiation.

Key Decisions for Investors

• No directional trade on the policy commentary alone; monitor formal U.S. agency action and state-level product-liability cases over the next 30-90 days before underwriting a regulatory multiple move.
• Build a 6-12 month watchlist for long PANW and CRWD versus a basket of unprofitable application-layer AI software names: liability-driven enterprise deployment should prioritize identity, monitoring, and incident response budgets. Enter only if bookings commentary begins explicitly citing AI governance demand; invalidate if net retention and billings fail to improve through two reporting periods.
• Prefer MSFT and AMZN to smaller model/application vendors on a 12-18 month horizon, as cloud providers can monetize compliance through managed AI, indemnification, and data-governance bundles. The key risk is aggressive liability pass-through to customers, which would slow workload consumption rather than enhance cloud share.
• Watch DDOG and ESTC for evidence that AI observability converts from experimentation to production spend. A material uplift in enterprise customers using governance/monitoring products, or management disclosure of AI-related expansion, would support a tactical long; absent such disclosure, treat the thesis as unconfirmed.