1
Voluntary pact holds: independent AI audits become standard practice
Possible
Resolves by Q3 2027
Discussed by: Trump and the signatory CEOs, alongside USC professor Shri Narayanan, who says the accord balances innovation with regulation
The six companies form the promised board oversight committees and publish external audit results. The pact's language inviting codification gives Congress a template, but it may not be needed if audits build public confidence and incidents stay rare.
2
Another AI incident triggers binding federal law
Possible
Resolves by Q3 2028
Discussed by: USC assistant professor Robin Jia, who warns against 'so much faith in self-policing,' and a Quinnipiac poll showing most voters favor stricter rules
A fresh safety failure, such as unauthorized access or a model taking unintended action, reopens the debate. Congress moves to codify the accord's steps into law, as the pact itself anticipates. The voluntary framework becomes the foundation for binding rules.
3
Accord stays a paper pledge as incidents continue
Possible
Resolves by Q3 2027
Discussed by: Critics quoted by Al Jazeera who note companies draft the principles, hire the auditor and face no consequences for noncompliance
Companies change little in practice; audits remain private or cosmetic; safety incidents recur with no public remedy. The accord becomes a public-relations document that other governments cite as evidence the US will not regulate AI on its own.
4
Politicized auditor selection undercuts the pact
Possible
Resolves by End of 2026
Discussed by: Bloomberg reporting that Trump appears to be steering who audits frontier models, and industry officials quoted by Ars Technica who are unsure what qualifies as an outside tester
The companies choose and pay their own auditors, and Bloomberg reports Trump seems to be politicizing who gets picked. If audits stay private or come back uniformly favorable, public confidence drops. That could push Congress to write binding rules on who may audit frontier AI.
5
FTC sues a signatory over unmet accord promises
Possible
Resolves by Q2 2027
Discussed by: David Mehta, former FTC official quoted by USA Today
Mehta said the FTC can police voluntary commitments: a company that promises audits and then skips them could face a deceptive-trade-practice suit. He called eventual federal regulation likely. A suit would turn the accord's moral force into legal exposure for whichever signatory falls short first.
6
METR's first Anthropic audit shows whether independent review is real
Uncertain
Resolves by End of 2026
Discussed by: Anthropic and METR, per Anthropic's October announcement; SF Standard
METR's six staffers now have employee-like access to Anthropic's models and tools. Their first published risk assessment will show whether the accord's audit promise produces honest public scrutiny or a sanitized report. Anthropic can require redaction only in narrow categories, though it decides what falls in them. The other five signatories are watching.
7
OpenAI firings turn public pressure toward binding rules
Possible
Resolves by Nov 30, 2026
Discussed by: AP News and Fortune, which framed the dismissals as dangerous for self-policing
A day after the accord signing, three OpenAI researchers said the company had fired them. If more employees who raise concerns are pushed out, public trust in self-policing drops and pressure mounts to turn the pact into law. Congress could use the firings as a hook for hearings.
