Anthropic is tracking which jobs are most exposed to AI. These 10 professions top the list.

Anthropic built an early-warning framework that compares AI capabilities to task mixes across occupations and identified the 10 U.S. professions most exposed to AI, led by computer programmers (75%), customer service reps (70%), data entry keyers (67%) and medical record specialists (67%), with financial & investment analysts at 57%. The company found limited evidence so far that AI has materially affected employment, though hiring of younger workers may be slowing in exposed roles, and BLS data cited by Anthropic project slower growth through 2034 for highly exposed occupations. For investors, the report flags medium- to longer-term sectoral and labor-cost risks for white‑collar, higher‑paid roles and potential repricing or operational shifts in companies heavily reliant on those workforces.

Analysis

Market structure: AI exposure (programmers 75%, customer service 70%, analysts ~57%) shifts value toward compute providers, model vendors and security tooling. Hyperscalers (AMZN, MSFT, GOOGL) and GPU leader NVDA capture both revenue and margin upside as firms buy inference + fine-tuning services; staffing/BPO and entry-level hiring are direct losers with demand down 5-15% in exposed roles in modelled scenarios over 1–3 years.

Risk assessment: Concentration risk (NVIDIA controlling >60–70% of high-end GPU supply) and regulatory tail risk (EU AI Act / US hearings in 6–18 months) are low-probability/high-impact events that could spike volatility and disrupt supply chains. Short-term (days–weeks) sentiment will move on layoffs/releases; medium-term (3–12 months) depends on model diffusion and corporate capex; long-term (1–5 years) is structural: wage compression in exposed jobs vs. higher demand for AI ops/cybersecurity.

Trade implications: Favor infrastructure and security exposure; avoid or short firms whose revenue is predominantly labor arbitrage (BPOs, staffing). Use size-constrained positions (2–4% per idea), horizon 3–12 months; options can monetize event risk around model releases and hyperscaler earnings. Watch compute pricing and NVDA earnings as 30–90 day catalysts.

Contrarian angles: Consensus underestimates secondary demand — model auditing, data labeling, and cyber-insurance could grow 20–40% y/y and benefit mid-cap SaaS/cyber names (CRWD, ZS) not yet fully priced. The market may also over-penalize legacy IT services; if regulatory action slows deployment, selective re-risking of large-cap tech (AMZN, MSFT) in 6–12 months will be attractive.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVIDIA (NVDA) over the next 2–6 weeks via staggered purchases; add a 1:1 call spread (3-month, buy ATM, sell +15% strike) ahead of the next earnings to cap cost while retaining upside to model-demand upside.
• Establish a 2% long position in Amazon (AMZN) focused on AWS exposure (buy shares or 6–9 month call spread) because hyperscalers will capture service revenue; size cautiously because near-term corporate cost cuts keep sentiment weak.
• Initiate a 1–1.5% short or buy 3–6 month put spread on a leading BPO/staffing firm (e.g., TTEC or MAN) sized to portfolio risk, targeting a 15–25% downside if hiring freezes widen; re-evaluate on quarterly hiring data.
• Overweight cyber-SaaS (examples: CRWD, ZS) by +2% portfolio weight via buys or 6–12 month call spreads, expecting 20–40% revenue growth in model-audit and monitoring services; take profits on +30% moves.
• Monitor EU AI Act progress and US congressional hearings over the next 30–90 days: if restrictive rules are proposed that materially limit model deployment, reduce NVDA/AMZN exposure by 25% and shift to defensive cash/corporates within 2 weeks.