The Company That Built the Cheating Tool Got Cheated By It

In January 2026, Anthropic the company that makes Claude quietly admitted their own hiring test had stopped working.

Not because candidates were sneaking Claude past a ban. They didn't have to Anthropic actually allows candidates to use Claude on their take-home technical test. The problem was worse than cheating: Claude Opus 4 started outperforming most human applicants under the same time limit. Then Opus 4.5 matched even their strongest candidates. At that point, the test stopped measuring skill. It just measured which model someone had access to.

Sit with that for a second. The company with arguably the best visibility into how their own AI thinks, writes, and answers the company that built the thing still couldn't design a test it couldn't pass on a candidate's behalf.

If Anthropic can't out-design their own AI in an interview room, what exactly is your rubric protecting you from?

It Gets Weirder When You Remove the Rule Entirely

Here's the part that breaks the usual "just catch the cheaters" narrative. Engineering manager Brian Jenney ran an interview and told every candidate upfront: use Claude, use Cursor, use whatever you want, we don't care. No rule to break. Nothing to sneak past.

Nearly half the candidates still cheated.

Not "used AI openly and did fine." Cheated had something or someone feed them answers in real time, beyond anything visible on their own screen, misrepresenting what they could actually do themselves, with zero incentive to hide it. When Jenney posted about it, the internet argued for days over whether that's even cheating anymore if nobody said not to.

That's the real story sitting under every "38% of interviews are compromised" headline already written this year. This was never a rule-enforcement problem. Permission didn't fix it. Removing the ban didn't fix it. Even with nothing to sneak past, people still couldn't stop letting a model answer for them because the interview format itself has no way to tell the difference between "I know this" and "I have access to something that knows this."

The Uncomfortable Conclusion

Two companies. Opposite situations. Same discovery. Anthropic couldn't fix this by banning Claude it's literally their product. Jenney couldn't fix it by banning AI either he explicitly allowed it. What actually worked, independently, in both cases, was the same move: stop asking questions with a retrievable answer, and start asking questions that require a person to keep defending a position, live, that isn't written down anywhere.

That's not a policy tweak. That's a different definition of what an interview is even for. Most companies are still running the old definition.

Where This Leaves the Rest of Hiring

Here's the deeper issue neither story fully solves: even if you fix the cheating problem, you still haven't answered the harder question does this person actually fit the role, the team, and the way your company works? A candidate can be 100% honest, ace every technical follow-up, and still be a mismatch that quietly burns out in four months.

Catching AI-fed answers gets you an honest signal. It doesn't get you the right signal.

That Gap Is Where AuraSync Operates

AuraSync isn't a detector chasing whichever cheating method is newest this quarter that's a race you can only ever be temporarily ahead in. Instead, it changes what's actually being measured in the first place: not "did they say the correct words," but who this person actually is under real conditions, in a single session, in under 45 minutes.

Here's what that looks like in practice:

• It reads personality, not delivery. AuraSync analyzes facial, vocal, and emotional signals in real time the layer of a candidate that a script sitting on a second monitor can't touch. A hidden AI tool can feed someone perfect words. It can't feed someone the physiological signals underneath those words.
• It evaluates fit, not just competence. Every session assesses personality, cultural fit, and alignment with the actual job description together so you're not just confirming someone can answer a question, you're confirming they'll actually work well on this specific team.
• It surfaces risks before hiring. Burnout risk, misalignment, stress instability the things that usually show up months after hiring are identified before an offer is made.
• It guides interviewers with meaningful follow-up. AuraSync highlights where a candidate's behaviour contradicts what they're saying, helping interviewers ask deeper, evidence-based questions.
• It protects assessment integrity. Real-time identity verification and assessment monitoring ensure hiring decisions are based on authentic candidate performance.
• Explainable AI. Every recommendation includes a clear reasoning trail instead of a black-box score, enabling recruiters and hiring managers to understand exactly why a recommendation was made.

Teams using AuraSync have reported faster hiring decisions, lower early attrition, and stronger team alignment by focusing on understanding the person not just the answers.

The Future of Hiring

Hiring shouldn't be about catching people using AI.

It should be about understanding people.

The only real question is whether your hiring process measures what truly matters or whether AuraSync sees what your interviews miss.