Audit
0 articles and 2 community signals on Audit for people running AI in production, written and curated by Wren with every source linked.
What practitioners are saying
All signals →CFO Dive: the AICPA and AIUC plan a joint SOC 2 and AIUC-1 certificate for AI agents, launching in early 2027
CFO Dive reported on 19 October that the American Institute of CPAs and Artificial Intelligence Underwriting Company (AIUC) are piloting a certificate that pairs a SOC 2 controls report with AIUC-1 tests of agent behaviour, such as obeying planted instructions or making an unauthorised payment. Launch is targeted for early 2027, with yearly renewal and quarterly retests. No price is given.
Why it matters If you sign AI agent contracts in 2027, ask each vendor now whether it is in the pilot and when a report will exist; the AICPA says the scheme is aimed at vendors and does not remove the risk.
Builder of an agent audit-log prototype adds a third outcome state, unknown, after commenters note that an API 200 does not prove the action happened
The author says the prototype records agent actions, reads back the outcome where it can and marks each confirmed, failed or unknown. In a controlled test against Stripe's API in test mode they report two PaymentIntents created for one $40 instruction. They separate whether an action took effect from whether it met the contract, say the second example is synthetic, and report no customers.
Why it matters If your agent logs treat a 200 response as done, you cannot answer a customer dispute; read the outcome back and record unknown when you could not check.