California AB 2013 Turns Training Data Into Public Mail

AB 2013 turns training-data questions into public statements

California AB 2013 requires developers of generative AI systems to publish documentation about training data characteristics. Read the bill text and history on California Legislative Information AB 2013. Governor Newsom’s 2024 signing batch shows how Sacramento is layering AI bills in a single season: Governor Newsom signing statement.

Reuters summarized the training-data transparency expectations for large developers: Reuters on California training data bill. NIST’s AI RMF remains the internal-control vocabulary engineers want beside legal drafts: NIST AI RMF.

If you cannot explain training data, customers will ask in email anyway

Transparency obligations become marketing and procurement artifacts. The failure mode is drafting too late—after comms promised something engineering cannot support.

A forwardable drafting loop

Audit SaaS Contract highlights risky terms in vendor and partner agreements. Email audit.saas.contract@via.email.

Summarize Contract Obligations extracts duties from DPAs and MSAs you attach. Email summarize.contract.obligations@via.email.

Generate Compliance Checklist builds a disclosure checklist from the statute language you paste. Email generate.compliance.checklist@via.email.

Assess AI Risk Exposure frames governance gaps from the narrative you supply. Email assess.ai.risk.exposure@via.email.

Distill to Three gives executives a three-bullet brief before publication. Email distill.to.three@via.email.

via.email does not publish your website. Counsel still approves final text.

Related reading

State AI rules stack fast. See EU AI Act Transparency Lands as Mail, Not Magic, Shadow AI Arrives by Forward. Governance Answers in Mail., and EU AI Act Meets Your Inbox Before Roadmaps. Agents at https://www.via.email/agents.

Engineering details vs. public language

The statute asks for documentation; the internet reads marketing. The failure mode is a public page that sounds confident while linking to nothing verifiable, or that exposes more than your security team wants public.

Run Assess AI Risk Exposure on an internal narrative draft before you polish the public version. Then have comms rewrite for readability. Then have counsel cut anything that overclaims.

AB 2355 is a reminder that email itself can be regulated speech

When political advertising rules intersect with synthetic media, teams discover that "disclosure" is not only a website footer problem. Keep an eye on cross-bill context as California keeps stacking obligations—your process should generalize, not whack-a-mole each statute.