Artificial intelligence is becoming a tool lawyers turn to for complex work and a sounding board for legal arguments. But a new report suggests that attorneys’ growing reliance on the technology is accompanied by a determination to check its output—and, for some, concern about their own job security.

Half of the lawyers surveyed for Bloomberg Law’s *AI in Practice* report said they trust AI tools enough to use them for complex tasks. More than 40% said they use AI to test legal arguments before taking them to colleagues.

That confidence has limits. Ninety-two percent of attorneys said they verify every citation and quotation produced by AI, underscoring the importance they place on checking the technology’s work even as they bring it into their practices.

The review process can also demand considerable time. More than a third of associates, 36%, said they spend more time correcting AI analysis than performing analytical tasks themselves. Among partners, 22% reported the same experience.

The findings portray a profession that sees practical uses for AI but remains alert to its shortcomings. Lawyers may be willing to enlist the tools for demanding assignments, yet many still devote substantial effort to examining and correcting what those tools produce.

The concerns extend beyond the quality of an individual answer. Three in four attorneys said they were at least somewhat concerned about AI agents acting outside their expected parameters, according to the report. Such concerns reflect unease about systems that may operate with greater autonomy than tools used to generate or test an argument.

The survey also found differences between associates and partners when it came to fears about employment. Thirty-six percent of associates said they feared losing their jobs to AI, compared with 16% of partners.

Together, the results suggest an uneasy relationship with a technology that lawyers are incorporating into their work while continuing to scrutinize its outputs and implications. For many attorneys, adoption has not displaced caution: trust in AI’s usefulness goes hand in hand with verification, correction and concern about what wider use could mean for the profession.