Quick Hits in AI News: AI Team Members, Shadow AI & More

Each week, as SHRM’s executive in residence for AI+HI, I scour the media landscape to bring you expert summaries of the biggest AI headlines — and what they mean for you and your business.

1. The Shadow AI Surge: Study Finds 50% of Workers Use Unapproved AI Tools 

What to Know: 

A new study by Software AG found that half of all employees use unapproved generative AI tools — like ChatGPT, DeepSeek, and Google Gemini — often without disclosing it. Workers are turning to artificial intelligence for tasks such as summarizing notes, writing emails, and coding. 

Convenience and speed are key drivers. However, this unsanctioned usage introduces real risks: 7% of users are already engaging with Chinese AI models, and sensitive data — including legal, financial, PII, and proprietary code — is routinely shared via unsecured platforms. Despite a slight drop in customer and employee data exposure, legal and code-related risks are rising.

Why It Matters: 

Shadow AI use is now mainstream — and it’s invisible to most HR leaders. This raises urgent issues around data privacy, compliance, and employee trust. HR must play a lead role in creating clear, safe policies that empower employees to use AI while protecting the organization.

2. Teach AI to Work Like a Member of Your Team 

What to Know: 

Generic AI tools often fail to deliver value because they don’t understand how teams actually work. At a Fortune 500 retailer, a contracts team saw no productivity gains from a standard large language model (LLM) until the company mapped the team’s real workflows into a “work graph.” This contextual data enabled a technique called reverse mechanistic localization (RML), which fine-tuned the AI based on how the team made decisions, retrieved data, and drafted contracts. The result: 50% less manual effort and a 30% increase in output.

Why It Matters: 

Off-the-shelf AI rarely fits real work needs because AI success requires tailoring tools to local team behavior — not just function. This means HR’s role in transformation is expanding — from rolling out tech to deeply understanding workflows, tribal knowledge, and user behavior.

3. Measuring Human Leadership Skills with AI Agents 

What to Know: 

A Harvard study shows we can measure leadership skills by seeing how folks manage GPT-4o simulated people (ρ = 0.81). In a randomized experiment, participants led both human and AI teams through complex group tasks. AI assessments strongly correlate (r = 0.81) with human team assessments. Success in both contexts was driven by the same core soft skills: the ability to ask the right questions, conversational turn-taking, emotional perceptiveness, and decision-making ability. Notably, demographic factors such as gender, age, and education had no predictive power. The research introduces AI-based leadership testing as a scalable, lower-cost way to measure leadership potential and soft skills.

Why It Matters: 

This opens a new frontier for leadership assessment. For HR, it offers a practical path to democratize access to leadership evaluation and refine talent pipelines. It’s interesting how management techniques apply to getting good work from both AI and humans, to a degree. Effective leaders ask questions, engage in conversational turn-taking, and have fluid and social intelligence.

4. AI and Jobs: Has the Inflection Point Arrived?

What to Know: 

This arXiv study tracks how AI tools like ChatGPT are reshaping global freelance markets — and finds that the impact varies dramatically by job. Translation freelancers saw declines in both volume and pay (displacement effect), while web developers experienced growth (productivity boost). The authors propose an “inflection point” theory: AI initially augments human labor — until it crosses a capability threshold, after which it begins to replace it. However, real-world shifts are slower than the hype suggests — translators, for example, earned more and saw higher demand a year after GPT-4 launched than during its release.

Why It Matters: 

Every role will eventually reach its own AI inflection point — so workforce planning must account for both short-term augmentation and long-term automation, with change unfolding more gradually than tech cycles.