LinkedIn will begin rolling out Hiring Assistant 2 in November, extending its AI recruiting agent beyond sourcing into screening and scheduling — and pairing it with new tools for flagging fraudulent candidate profiles.

The company announced the update Wednesday at its Talent Connect conference in New York. Existing Hiring Assistant customers using the product in English receive it automatically, at no added cost. The release touches two decisions HR teams now face directly: how much of candidate evaluation to hand to an agent, and how to verify identity as fraud concerns grow.

The commercial stakes are clear. LinkedIn said its agentic hiring products have passed a $450 million annualized revenue run-rate, with more than 20,000 companies using them. The products helped LinkedIn grow revenue 12% in the April-to-June quarter even as overall hiring fell.

Candid about version one

LinkedIn CEO Dan Shapero and Vice President of Product Management Dan Reid acknowledged in their keynotes that feedback on the first Hiring Assistant was mixed. Shapero relayed an internal anecdote about a product team shipping a new model.

"What they claimed was that in the last four weeks since we shipped it we've iterated, and now the expletives in the product from recruiters have gone down," Shapero said.

Customer feedback drove the rewrite. As Reid put it, customers wanted Hiring Assistant to "just be smarter."

What changed under the hood

Hiring Assistant 2 is contextually aware, carries a persistent memory tied to each recruiter, and makes inferences that respond to user intent. The design goal, per LinkedIn: do what the recruiter meant, not what they said.

In practice, a sparse search now fills in context from the recruiter's and company's hiring history. LinkedIn's example: a request for "product manager" returns candidates near the office, in retail and at mid-level, because that is who the recruiter has hired before. The system can also apply company rules, such as never recruiting from the company's own subsidiaries.

Recruiters can type criteria that don't function as keywords — "fresh perspective," for instance — and the system interprets them. Recruiters can review and confirm the inferred criteria before results are locked in.

Each recommended candidate now comes with two explanatory signals. Fit signals give a personalized read on role match and can flag mismatches, such as a candidate who grows existing accounts when the role calls for a prospector. Candidate interest estimates how receptive someone is likely to be, drawing on platform engagement, Open to Work details and InMail response frequency.

Fraud takes center stage

The second driver was candidate fraud. LinkedIn's internal research found that 64% of US talent acquisition professionals say it is getting harder to tell who and what is real.

"Ten years ago, this was not the problem we were talking about," Shapero said of verifying candidates' identity. "But this is the problem we have today."

LinkedIn says 115 million members have verified their identity to date, and announced three changes. Employers will be able to flag profiles that falsely claim to have worked at their company and disconnect them from the company page. Verification will be built into the job application itself. And Connected Apps — launched in June, linking a member's profile to tools they use and displaying what they have done in them — will expand the proof-of-skills pool. The Connected Apps menu currently lists 26 partners; ElevenLabs and Medium, announced during the keynote, do not appear there yet.

Hiring Assistant 2 will combine these validations with ATS data, pre-screening answers and work history to build a fuller candidate portrait and flag inconsistencies.

Deeper ATS integration

The first Hiring Assistant mostly handled the front end of recruiting: intake, sourcing and outreach. Version 2 reaches into screening and scheduling.

Companies that connect their applicant tracking system can have the agent evaluate their entire talent pool, including candidates who applied outside LinkedIn. At launch, it integrates with Greenhouse, Lever, SmartRecruiters, Zoho Recruit and Tracker. When a recruiter moves a candidate forward in LinkedIn, the change now syncs back to the ATS.

LinkedIn says Hiring Assistant completes pre-screening in a median of six minutes and can decide what happens next for each candidate. A standout might be flagged for the recruiter to contact personally; borderline candidates go to a screen.

For talent acquisition leaders, the adoption questions are now concrete. Does inference from hiring history bake past biases into sourcing? Who reviews the inferred criteria before they shape a pipeline? And how much screening authority should sit with an agent before a human sees the candidate?

Data in, data out

Nearly everything new in Hiring Assistant 2 depends on more data flowing into LinkedIn: colleagues vouching for one another, employers flagging fake employees, ATS vendors syncing applicant pools and app partners reporting what members do on their platforms.

At the same time, LinkedIn is restricting what flows out. In September it took down the company pages of Pin and Gem, two companies with AI sourcing products. Pin says it is still operating and working with LinkedIn to resolve the issue. A LinkedIn spokesperson said: "LinkedIn takes action when necessary to prevent our members' information from being scraped and used without their consent."

In the age of AI, everyone is consolidating data to feed the models — and as the Pin and Gem removals show, LinkedIn intends to control whose models get fed.