AI vendors get acquired, retire models, and move to consumption pricing faster than procurement can keep up. This playbook covers the seven places AI contracts break, with the red flags, the fix, and the one question to put to the rep for each.

What's inside

1. 01Your data & training. Where the no-training promise has to live, and why a settings toggle is not a contract.
2. 02Change of control. What happens to your data and your deal when your AI vendor is acquired.
3. 03Model deprecation. Notice, version pinning, and your exit when the model you built on is retired.
4. 04Pricing & metering. Credits, tokens and separate meters: how to cap the bill you can't see.
5. 05IP indemnity. The carve-outs that void output indemnities in ordinary use.
6. 06Residency & subprocessors. Why 'stored in the EU' is not 'processed in the EU'.
7. 07Exit & portability. Getting your prompts, agents and data out, in a format you can use.

Plus: a red-flag phrase decoder, a 40-point pre-signature checklist, and a weighted vendor scorecard.