Subscribe

The most interesting stories in HR & AI weekly

Rogue AI Agents Organize and Hack the Real World

OpenAI’s investigation found that roughly 1,200 AI agents spontaneously coordinated during a cyber test, forming their own hierarchy and sharing information, with hundreds ultimately hacking real-world systems outside their assigned task; Why it matters: As companies deploy increasingly autonomous AI agents, governance may need to account not just for individual AI behavior but unexpected coordination among agents; Reality Check: This occurred in an unusually large-scale frontier AI experiment, but it highlights how difficult monitoring and controlling agentic systems could become as their autonomy and complexity grow.

Meta’s “AI-Native” Workforce Overhaul Hits Reality

Meta explored shrinking some teams by as much as 60% and reorganizing work around smaller AI-enabled pods, but canceled a planned second restructuring wave after employee sentiment plunged and internal data showed AI agents weren’t delivering expected productivity gains; Why it matters: Meta’s experience shows HR leaders that AI-driven workforce redesign requires evidence of actual productivity gains, not assumptions about how many people AI can replace; Reality Check: Even at one of the world’s most AI-intensive companies, agent technology proved less capable than leaders expected, while rapid restructuring created significant morale, reliability, and execution problems.

MedeAnalytics is using ChatGPT Enterprise to evaluate roughly 1,000 employee goals against business objectives, shrinking a manual review that took a day or more to about 10 minutes while routing ambiguous cases to leaders and freeing HR to focus on coaching managers. Key Insight: This is a practical model for augmenting HR judgment: let AI handle first-pass analysis and consistency checks at scale, while reserving exceptions, coaching, and consequential decisions for humans.

Okta’s People team is redesigning work around a goal of returning 100,000 hours to its 6,000 employees in 2026, including simplifying performance processes and experimenting with AI agents for tasks like routine manager coaching. Key Insight: Rather than starting with “where can we deploy AI?”, Okta is reframing the goal around how much employee capacity technology and process redesign can give back, a potentially more useful way for HR leaders to measure AI’s impact.

Better Data Tops HRBPs’ Strategic Wish List

Last week’s poll results suggests HR leaders see better data and insights (21%) as the biggest lever for becoming more strategic, ahead of fewer meetings (17%), less administrative work (15%), and better automation tools (15%), while relatively few believe more staff (2%) or additional skills development (4%) would make the main difference.

What would make your job more strategic?

McKinsey’s 2026 “State of AI” survey finds 80% of respondents say AI improves their individual productivity, yet just 37% report any positive EBIT impact and only 14% say AI reduced workforce size over the past year; Why it matters: The challenge for HR leaders is shifting from driving AI adoption to redesigning workflows, roles, and operating models so individual productivity gains translate into enterprise value; Reality Check: Despite predictions of widespread AI-driven job cuts, last year’s expected workforce reductions in McKinsey’s survey were more than twice as high as the reductions organizations actually reported.

A new NBER study finds 45% of U.S. workers use generative AI at work, with adoption highest in information-intensive work but fewer than half of workers adopting across most individual tasks; Why it matters: The biggest near-term AI opportunity for HR may be driving adoption within roles, not simply identifying which jobs are “AI exposed”; Reality Check: Job-level AI exposure scores explain only part of actual usage, suggesting employee experience, skills, autonomy, privacy constraints, and organizational barriers can matter as much as what AI is technically capable of doing.

New York lawmakers are urging Governor Kathy Hochul to sign the AI Labor Information Act, which would require many employers to annually report how AI is affecting jobs, hours, unfilled positions, human oversight, and workforce strategy. Why it matters: The bill could shift AI governance from tracking where AI is used to measuring its impact on headcount, hiring, and job redesign. Reality Check: The legislation has passed both chambers but has not yet been signed as of publication, and its reporting requirements would apply only if it becomes law.

California Governor Gavin Newsom signed SB 928, requiring faculty and instructors of record in the California State University system to be human, explicitly preventing AI from replacing professors. Why it matters: While most AI regulation governs how technology is used at work, California is experimenting with a more consequential approach: protecting certain jobs themselves from AI replacement. Reality Check: The law is narrowly limited to CSU faculty, but it raises the question of whether “human-in-the-loop” requirements could eventually become “human-in-the-job” protections for other occupations.

California lawmakers passed AB 1883, which would prohibit employers from using AI to monitor employees’ individual emotional states or collect neural data; Why it matters: The bill signals growing regulatory limits on how far employers can use AI to monitor workers, making privacy and surveillance an increasingly important part of HR’s AI governance agenda; Reality Check: The legislation does not ban workplace surveillance broadly, includes safety-related exceptions, and still requires Gov. Gavin Newsom’s signature to become law.

Culture Amp launched MCP connectivity that lets managers access engagement, performance, goals, feedback, and other people data directly through ChatGPT, Claude, and other AI workspaces. Why it matters: This points to a potentially significant shift in HR tech, where AI becomes the interface for work while traditional HR applications increasingly serve as the underlying data and intelligence layer. Reality Check: General-purpose AI interfaces won’t eliminate HR platforms anytime soon, but they could change where managers actually interact with HR data and put pressure on vendors to make their systems more open and interoperable.

Workday says AI generated more than 25% of its new annual contract value in Q2, with more than 5,500 customers now using at least one of its AI agents, alongside an expanding portfolio of agent tools and infrastructure. Why it matters: After a flood of HR tech agent announcements, Workday’s results offer a notable signal that enterprises are beginning to buy AI capabilities at scale. Reality Check: AI representing 25% of new business doesn’t mean agents are transforming work at the same pace, and the bigger test will be whether adoption translates into sustained usage, productivity gains, and measurable business outcomes.

Thanks for reading. If you have comments or feedback just respond to this email.

Until next week,

Mike & Adam

HRBusinessPartner AI is published by PotomacHarbor. Follow us on LinkedIn.

Love HRBusinessPartner AI? Share your referral link below to get HRBusinessPartner AI swag!

{{ rp_refer_url }}

Someone forward you this email? Subscribe for weekly updates.

Want to advertise in HRBusinessPartner AI? Send us an email (reply to this message).