Careers Briefing — September 10, 2026

Three of today's biggest careers-and-AI stories point in the same direction: the guardrails around AI are being drawn in public — by researchers, regulators and intelligence agencies — and each one is a distinct hiring signal for job seekers weighing where to specialize. A frontier lab is now publishing extinction-risk odds and staffing safety headcount around them. California is moving to legally define what employers can measure about a worker's nervous system. And the US is treating large-scale model distillation as a national-security matter, opening a new compliance surface for the industry.

Is Anthropic putting real money behind its extinction warnings?

Yes — the researcher who publicly attached a "greater than 10 percent" figure to AI killing all humans this decade works at a lab that is actively hiring 42 Safeguards roles plus a set of alignment and interpretability positions built around exactly that risk. Evan Hubinger, who leads Anthropic's alignment science team, made the estimate this week in response to former Anthropic and OpenAI researcher Jacob Coxon, who quit and posted, "Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives," according to CBS News and Fox Business. Hubinger replied that Anthropic is trying its best but does not "yet have a plan to solve alignment for superintelligence." Anthropic's own careers page shows 66 open roles under AI Research & Engineering — including interpretability and alignment scientists — and 42 open Safeguards positions covering biological safety, enforcement and policy design. The signal for job seekers: safety and alignment work is now funded core headcount at a frontier lab, split into two distinct job families (research-track alignment/interpretability and trust-and-safety/policy) with different skill demands.

What would California's emotion-AI ban actually make illegal at work?

California lawmakers passed Assembly Bill 1883, which would prohibit employers from using AI to detect or infer employees' emotional states, and the bill now awaits Gov. Gavin Newsom's signature, according to HR Executive. The bill bars collecting "neural data" — information generated by measuring activity in a worker's central or peripheral nervous system, "not inferred from nonneural information" — and targets AI layered onto existing surveillance systems such as video, audio, time-tracking and geolocation tools. Violations would be enforced by the California Labor Commissioner and public prosecutors, with civil penalties of up to $500 per violation, per HR Dive. Exemptions cover safety-purpose surveillance and operations tied to aircraft, national security, military, space and defense. Standard security systems, facial recognition and safety analytics — like detecting a fatigued driver — would still be allowed because the bill targets emotion inference specifically. The EU's AI Act already bars similar workplace emotion-recognition tools; Korn Ferry, citing an industry report, has said the emotion-AI market could reach $9 billion by 2030. Ask the question regardless of where the job is based — most states have no equivalent restriction on the record yet.

Is the US calling model distillation a security issue — and what does that mean for AI-policy jobs?

Yes. The NSA, FBI and CISA issued a joint advisory on Sept. 9 accusing DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI of "aggressive, malicious, and targeted distillation activities at an industrial scale," according to The Hacker News. The agencies said the firms extracted "billions of tokens across millions of exchanges/requests" from Anthropic's Claude, OpenAI's GPT, Google's Gemini and xAI's Grok since at least late 2024. Google's own Threat Intelligence Group separately said this week that some individual distillation campaigns against its models exceeded 100 million prompts. Engadget's reporting adds specifics: DeepSeek allegedly used data from Claude, Gemini, GPT and Grok to help train its R1 reasoning model, while Moonshot AI allegedly drew on Claude data for Kimi K3 and GPT-4o data for Kimi K2. Alleged methods span VPNs, obfuscated accounts, third-party API aggregators, bulk-bought premium subscriptions shared across developer teams, and gray-market account proxies sold on Taobao and Xianyu. The signal: distillation is a legitimate ML technique — the advisory targets scale and method, not the practice itself — and elevating it to a joint federal advisory turns AI-policy, export-compliance, threat-intelligence and China-desk expertise into concrete hiring lanes at both AI companies and the firms that serve them.

What to watch

• Whether Gov. Newsom signs AB 1883 into law — the bill's fate sets the near-term US floor for employer emotion-inference tools.
• Anthropic's next 30 days of Safeguards and alignment hiring pace, and whether other frontier labs match the public risk framing with comparable headcount.
• How AI companies respond to the joint federal advisory operationally — expect more visible detection-and-response investments and new client-facing compliance requirements downstream to enterprise buyers.

Sources

• Anthropic researcher says more than 10% chance AI could kill all humans — CBS News
• Anthropic researcher says AI has over 10% chance to 'kill all humans' within next decade — Fox Business
• Who Is Jacob Coxon? Anthropic Researcher Quits — Warns AI Could Kill Everyone — Newsweek
• Anthropic Careers — Anthropic
• California bill would ban AI tools that read employees' emotions — HR Executive
• California ban on workplace AI emotion surveillance heads to Newsom — HR Dive
• US Agencies Accuse China AI Firms of Industrial-Scale Distillation — The Hacker News
• US authorities accuse Chinese AI companies of industrial-scale campaigns to copy American models — Engadget