There is growing momentum among industry leaders like OpenAI, Anthropic, and Google DeepMind to explore a shared AI safety standards framework that would coordinate testing and disclosure across frontier models, signaling a rare cooperative push among rivals.

Proponents argue the framework can be put in place quickly and without a UN-style regulator, drawing a parallel to self-regulation seen in the film industry.

Implementation would need to tackle IP and US-China regulatory gaps, with ideas like encrypted testing environments and third-party oversight proposed to ensure compliance.

While the plan isn’t perfect, supporters say external scrutiny could substantially improve safety by increasing the chances of uncovering problems.

The core concept is a self-disciplined, industry-led defense line: competitors would share APIs, conduct independent security testing, and use shared logs and open-source tools to prevent information leakage.

To avoid biased self-assessment, the approach calls for diverse testing from multiple companies to prevent overfitting to specific benchmarks.

Cross-testing is seen as potentially better at spotting risky behaviors before release, though it raises concerns about IP protection and security.

The discussions come amid a broader debate about whether to accelerate development or prioritize safety, with some leaders advocating slower progress or more cautious governance.

Although peer testing is unusual in this fast-moving field, it could influence how companies balance safety with commercial and national-security considerations.

Musk and others stress external evaluation’s urgency, acknowledging concerns about information leakage while pushing for rapid safety address.

Competitors would flag safety concerns and raise alarms if dangerous behavior is detected, rather than each lab auditing its own model.

A future standards body could set technical safety requirements while preserving competitive dynamics in product areas outside safety.