Thinkers50: 74% give AI agents access; 78% doubt audits

Thinkers50 reported on 2026-10-07 that a 2026 Grant Thornton survey of 950 executives found 74% let AI agents access data and business processes, 78% do not trust their organizations to pass an independent AI governance audit within 90 days, and only 20% have a tested contingency plan. In the same essay, Thinkers50 presents Hamilton Mann’s argument that leaders should legitimate algorithmic power through a separation‑of‑powers model.

According to Thinkers50, the survey describes exposure in mid‑size and large organizations that already allow software agents to interact with operational data and workflows. Thinkers50 characterizes Mann’s proposal as checks and balances that distinguish those who build, those who deploy, and those who review and sanction decisions.

Thinkers50 cites recent cases to illustrate risk. It reports that in 2019 Deanna Amato discovered the Australian government owed her thousands of dollars after seizing her tax refund generated by the Robodebt program. It also reports that a Chinese national, Senbin Hou, was detained at an ICE office near Seattle in May 2026 after a face recognition system registered 15 failed biometric check‑ins, and that a federal judge ordered his immediate release in July 2026, finding the detention violated due process.

What it means

If your teams use AI agents in production, ask how they would evidence audit readiness within 90 days and when the contingency plan was last tested, as Thinkers50 reports many lack both.