OpenAI and Anthropic reportedly abandoned mutual safety testing deal

The article argues that voluntary cross-testing is inherently unstable because it requires one lab to give a rival privileged access to probe its model. This involves disclosing capabilities, guardrails, and failure modes, which are typically protected by trade secret law. While the safety benefits are diffuse, the commercial cost of revealing where a model breaks is immediate. Consequently, the arrangement is unlikely to remain voluntary without external mandates.

Microsoft, which holds a roughly 27% stake in OpenAI valued at about $135 billion and booked a $3.2 billion gain on its Anthropic investment in fiscal Q4 2026, faces governance and liability risks. If regulators mandate external AI model validation, Microsoft’s position with Azure’s $100 billion revenue and 30 million Copilot seats could allow it to profit most, but the current lack of agreement shifts risk onto its balance sheet.