TalkOnPointAlex Bores’s proposed safeguards for catastrophic AI risks
What AI regulations does Alex Bores propose? Alex Bores proposes requiring third-party audits of all frontier AI labs, ideally allowing outside evaluators employee-level access to assess systems directly ([13:40.696](#)). He also proposes monitoring every AI deployment, including evaluations, and requiring retention of AI logs so past actions can be reviewed ([13:55.196](#); [14:10.376](#)).
Direct answer
Alex Bores proposes requiring third-party audits of frontier AI labs, monitoring every AI deployment, retaining AI logs, and mandating safety-incident reporting with underlying data and other metrics. He also supports changing broader liability rules and increasing companies’ liability, but the supplied evidence does not specify a complete allocation of responsibility. He suggests considering restrictions on superintelligence or recursive self-improvement while calling the other measures basic safeguards. Third-party auditors should be independent, have employee-level access, and operate with government or certified oversight; the source does not establish that every proposal is enacted.
Top points
Bores proposes third-party audits of all frontier AI labs, ideally with employee-level access for outside evaluators. [13: 40.696]
He proposes monitoring every AI deployment, including evaluations. [13: 55.196]
He proposes retaining AI logs so past actions can be reviewed and customers cannot delete them. [14: 10.376]
Main points
Choose a numbered topic, move with Previous and Next, or play the exact evidence in the source video above.
Bores proposes third-party audits of all frontier AI labs, ideally with employee-level access for outside evaluators. [13: 40.696]
He proposes monitoring every AI deployment, including evaluations. [13: 55.196]
He proposes retaining AI logs so past actions can be reviewed and customers cannot delete them. [14: 10.376]
He proposes mandatory reporting of safety incidents, including underlying data, chain-of-thought information, and other metrics. [14: 25.216]
He says broader liability rules should change and suggests considering restrictions on superintelligence or recursive self-improvement. [14: 40.316]
He supports government involvement and certified independent verification organizations for audits, with access and reporting protections intended to preserve independence. [38: 50.432]
Structured summary
The complete public summary remains crawlable and linked to the original source.
What AI regulations does Alex Bores propose?
Alex Bores proposes requiring third-party audits of all frontier AI labs, ideally allowing outside evaluators employee-level access to assess systems directly ([13:40.696](#)). He also proposes monitoring every AI deployment, including evaluations, and requiring retention of AI logs so past actions can be reviewed ([13:55.196](#); [14:10.376](#)).
He further proposes mandatory reporting of safety incidents, including the underlying data, chain-of-thought information, and other relevant metrics ([14:25.216](#)). He describes these measures as basic safeguards and says broader liability rules should change.
He also suggests considering restrictions on superintelligence or recursive self-improvement ([14:40.316](#)).
Who should be held accountable for AI risks?
Bores supports changing broader liability rules and increasing companies’ liability; he says he proposed a New York bill for that purpose ([32:27.220](#)). However, the supplied evidence does not specify a complete allocation of responsibility among particular people or organizations.
What role would third-party audits play?
Bores favors strong government involvement in auditing, certified independent verification organizations, employee-level data access, and reports that companies cannot censor. Ideally, auditors would neither be paid nor selected by the companies they review ([38:50.432](#)).
New York’s RAISE Act, as described in the source, requires large frontier-model developers to establish and disclose safety and security protocols ([5:30.584](#)).
TalkOnPoint used AI to organize the source into a readable summary and connect important topics to supporting source moments. This analysis may contain errors; use the cited excerpts, timestamps, and original source to verify consequential information.