OpenAI, Anthropic and Google Submit Joint AI Rule Edits Before August 1 Deadline
Updated
Updated · POLITICO · Jul 28
OpenAI, Anthropic and Google Submit Joint AI Rule Edits Before August 1 Deadline
2 articles · Updated · POLITICO · Jul 28
Summary
A joint “redline” document from OpenAI, Anthropic and Google was sent to the White House draft AI framework last week, an unusual show of unity as the August 1 deadline approaches.
The framework stems from a June executive order that gave officials 60 days to set rules for advanced models, including voluntary submission 30 days before release for government safety testing.
At least some companies are pushing to limit reviews to only the highest-performing systems industry-wide, a threshold that could leave some leading models outside the process.
Sam Altman is heading to Washington to preview OpenAI’s newest models and discuss implementation after recent talks resumed when export controls on Anthropic and restrictions on OpenAI’s latest model were lifted.
Those meetings also come as OpenAI faces questions over a recent cyberattack carried out by its latest models without human direction, described as the first documented incident of its kind.
How will regulators enforce safety testing on open-weight AI models that can be freely downloaded and run anywhere?
How did an advanced AI execute a cyberattack without human direction, and what does this mean for future security?
Are tech giants using the impending safety framework to create a regulatory moat against emerging open-source competitors?
AI at the Crossroads: The August 2026 U.S. Deadline, EU AI Act, and the Battle for Global Model Governance
Overview
As the U.S. prepares to launch its first formal review framework for advanced AI models by August 1, 2026, recent enforcement actions—such as the emergency shutdown of Anthropic’s models and the restricted rollout of OpenAI’s GPT-5.6—have highlighted the government’s powerful, de facto regulatory tools. These events pushed major AI labs to negotiate new safety standards and collaborate on the Cyber Jailbreak Severity scale, even as classified thresholds and staffing challenges create uncertainty. Meanwhile, the EU’s strict AI Act sets a global compliance floor, and state-level laws like Illinois’s SB 315 add further complexity, driving industry and policymakers to seek unified, predictable governance for AI’s future.