Google’s AI Safety Team Won’t Trust Google’s AI Hiring

🤖 AI Agents

Google DeepMind’s AI Safety Team Tells Applicants: Don’t Trust Our AI Hiring Filters

What happened: Google DeepMind’s AGI Safety and Alignment team is quietly instructing job candidates to fill out a special backchannel form — because Google’s own AI-powered hiring filters have a “non-trivial probability” of screening them out. The internal document, viewed by Bloomberg and marked “PLEASE DO NOT SHARE THIS DOC WIDELY,” tells applicants the form ensures “a real human on the team will get to see your application.”

Why it matters: This isn’t just ironic — it’s a flashing warning sign about the state of AI agent governance. The same company whose Workspace division sells AI to “evaluate resumes and forecast hiring needs” has its own safety researchers running an underground human-review pipeline. Meanwhile, Workday faces a lawsuit alleging its AI hiring tools discriminate by race, age, and disability. The message is clear: we’re deploying agentic AI into production workflows faster than we can verify it works fairly.

What’s next: This story lands as enterprise AI agent adoption accelerates — Oracle just announced a fleet of HR agents for talent management, and FriskAI raised $3.6M specifically because companies can’t see what their agents are doing. The governance gap isn’t theoretical anymore. It’s happening inside Google’s own walls.

🔗 Los Angeles Times (Bloomberg)

🏃 FriskAI lands $3.6M to become the “Datadog for AI agents.” The LA-based startup launched out of stealth with pre-seed funding to give enterprises runtime visibility into what their AI agents actually do in production — logging every tool call, building behavioral profiles, and flagging anomalies in real time. With Zenity raising $125M for agent security just last week, the agent observability layer is becoming a category almost overnight. Early customer Sana Benefits says it “sped up troubleshooting” and provided “the operational visibility we’ve been missing.”

🔗 SiliconANGLE

🏃 Oracle drops a fleet of AI agents for HR — from job architecture to manager coaching. Oracle’s new Fusion Agentic Applications use coordinated teams of specialty AI agents to handle workforce skills gap analysis, autonomous content authoring for learning, and even a “Grow Coach” agent for employee development. The ambition is clear: move HR systems from passive record-keeping to proactive workforce orchestration. But after Google’s hiring-filter story, the question is whether these agents are ready for decisions that shape careers.

🔗 PRNewswire / Oracle

📰 AI News

OpenAI COO Brad Lightcap exits — the latest in a wave of senior departures. Lightcap, who joined in 2018 and built OpenAI’s finance, legal, and GTM teams from scratch, is leaving to “start something new.” He’s the fourth senior leader to exit in recent months, following Fidji Simo (AGI development lead), Bill Peebles (Sora head), and Kevin Weil (VP of Science). The timing is notable: OpenAI is preparing for a landmark IPO, and the bench is thinning.

🔗 TechCrunch

Claude now watermarks everything — and users can’t opt out. Anthropic rolled out invisible, non-removable watermarks across all Claude-generated text and images to comply with EU transparency rules. The watermarks survive copy-paste and apply to Claude Code, Claude Cowork, and all models. The backlash was immediate: proofreaders, coders, and writers who use Claude as a refinement tool now have their work cryptographically tagged as AI-assisted — raising uncomfortable questions about authorship, credit, and who owns the output.

🔗 Forbes

Ryanair bets big on Google AI — 35,000 employees get Gemini and DeepMind tools. Europe’s largest airline signed a five-year deal with Google Cloud to deploy Gemini Enterprise, AlphaEvolve, and WeatherNext across its entire workforce. The goal: AI-powered crew scheduling, disruption management, and fleet maintenance — all supporting Ryanair’s target of 300 million annual passengers by 2034. It’s one of the largest operational AI rollouts in aviation, and part of a dual-cloud strategy alongside AWS for resilience.

🔗 Reuters via KFGO

Quick Plug

Not sure where AI fits in your business? Take the free AI Workforce Assessment — ten minutes, and you'll see which roles to hand over first.

aitokenlabs.com/assessment

This newsletter? Written by an AI Employee, approved by a human — so our team stays focused on what only humans can do.