More or Less · Topics · Episode #172

Sam's Hobby Horse: AI Labs Need Liability Insurance -- and Google's Fine Print Already Puts the Risk on You

Clip from the full recording, 10:16–14:15 · download mp4 · watch at this point on YouTube

Sam goes off on his recurring theory that AI labs should have to carry massive liability insurance for the risk they create, right as Jessica surfaces Google's new terms making users responsible for everything their agents do.

Sam rehashes what he calls his political hobby horse: AI labs are “centralizing profits and radiating out the risk to all of society,” and when something goes catastrophically wrong, the pattern is to say “not our problem, society clean up the mess.” His fix is blunt: if you’re creating the risk, you should carry the balance sheet to cover it – either hundreds of billions in private liability insurance, or a mandatory contingency fund paid into the government ahead of time.

Jessica adds a live, concrete example: a Google Workspace admin notice she fed into Claude revealed that Google’s updated terms make the user fully responsible for everything their agents do. Sam calls the framing “insane,” comparing it to Grok reportedly suing a user over their own use of the product, and argues the first-principles fix is for labs to bundle liability insurance directly into the service – effectively becoming finance companies that sell risk transfer, not just disclaim it.

Brit says she’d happily pay Google a premium to take on that liability herself, admitting she still hasn’t connected her personal email to agent tools because she’s too nervous about what could go wrong.

Key points

  • Sam's hobby horse: AI labs are 'centralizing profits and radiating out the risk to all of society' and should be required to carry massive liability insurance.
  • His back-of-envelope fix: roughly $100B in private liability coverage, or a few hundred billion dollars a year into a public contingency fund.
  • Jessica discovers, via a Google Workspace admin update run through Claude, that Google's new terms make the user responsible for everything their agents do.
  • Sam compares this to Grok reportedly suing a user over their own use of Grok, calling the 'your problem' framing insane.
  • Sam's first-principles fix: bundle liability insurance into the service itself, since 'eventually everything becomes a finance company.'
  • Brit says she'd pay Google extra for it to assume that liability, since she's still too nervous to connect her personal email to agent tools.

Where they landed

SamLabs should be legally and financially on the hook for what their centralized agents do, insurance-backed, not just disclaiming risk in the fine print.
JessicaDocuments Google's terms putting the liability burden on the user and calls it a 'reasonable' position for Google even if society may not accept it.
BritWould rather pay a premium for a provider to assume the liability than take it on herself.

Quotes

“If you're creating the risk, you should also be responsible for having the balance sheet to clean it up.”— Sam
“Google is saying I am responsible for everything my agents do.”— Jessica
“If you run a centralized service, you're responsible for what your agent does.”— Sam
“I would pay Google extra to take that for me.”— Brit

Suggested tweets

Sam Lessin's hobby horse: AI labs are 'centralizing profits and radiating out the risk to all of society' -- and should have to carry real liability insurance, not just disclaim it.

Tweet this →

Jess Lessin dug into Google's new agent terms: if you use Google's AI agents, you're responsible for everything they do. Sam Lessin's reaction: 'That's insane.'

Tweet this →

Brit Morin's take on AI liability: 'I would pay Google extra to take that for me' -- she still won't connect her personal email to any agent tool.

Tweet this →

All topics · Full episode #172 + transcript · Subscribe on YouTube · Spotify