By Lee Flanagan
✨ AI Summary:
- AI screening tools that require insider knowledge to navigate select for access and connections, not actual ability or fit.
- When your own technical experts coach candidates on how to game your hiring filters, the system has lost credibility and is measuring the wrong things.
- Candidate opt-out rates on AI interviews reveal trust gaps that scheduling flexibility cannot solve—the issue is relational, not logistical.
- Technology adoption in hiring fails when trust in the tool lags behind; audit whether candidates and hiring teams actually believe your system makes fair decisions.
Weekly Roundup · 2026-08-10 to 2026-08-14
Google DeepMind’s AGI Safety and Alignment Team built a workaround for its own hiring filter. The team developing artificial general intelligence coached candidates on how to get past the company’s AI screening tool. If the people building the future of AI do not trust AI to hire them, the rest of us should pay attention.
The team building AGI does not trust AI to screen its own candidates
When your most technical team tells candidates how to game your hiring filter, you have a signal problem, not a candidate problem. The irony is sharp, but the lesson is broader: if the system requires insider knowledge to navigate, it is selecting for access, not ability. This pattern appears everywhere. The tools meant to surface talent instead obscure it, and the people closest to the work know it first.
Flexibility is not the same as trust
Late-night AI interviews are sold as candidate-friendly scheduling. Greenhouse’s data shows candidates opting out in meaningful numbers. The pitch assumes the friction is logistical. The reality is relational. A 1am slot does not fix the fact that candidates do not trust a machine to understand their answers, and no amount of convenience will close that gap.
Two stories, one pattern: the technology is ahead of the relationship. When your own experts route around the tools and your candidates opt out of the convenience, the problem is not adoption. It is trust. And trust is not a feature you can ship.
