AI Hiring Brief · 2026-07-09

19 frontier AI labs · 84 roles added in 24h · agihunt.info/en/jobs

Today at a glance

Anthropic's threat-intelligence buildout: Three "Threat Intel Manager" roles — covering "CBRN-E & Advanced Weapons," "Model Exploitation & Fraud," and "Influence Operations & Surveillance" — landed simultaneously alongside a "Product Manager, Safeguards (Child Safety)." Nearly half of Anthropic's 9 additions today were in the Safeguards (Trust & Safety) department, an unusually concentrated single-day signal.
OpenAI pushes into Asia-Pacific commerce: "AI Deployment Engineer, Cyber" appeared in both Tokyo and Singapore; "GTM Strategy & Operations, India & South East Asia" landed in Singapore; and "Executive Business Partner, Korea" went up for Seoul — four Asia-Pacific commercial postings in one day spanning Japan, Singapore, and South Korea.
NVIDIA concentrates on autonomous-vehicle perception: Three separate "Senior Perception Engineer - Autonomous Vehicles" listings (all Santa Clara) plus a "Senior Software Engineer, World Model Systems Engineering - Autonomous Vehicles" made AV the most focused theme in today's NVIDIA additions, with four of eleven new roles tied directly to the autonomous-vehicle stack.
SpaceXAI builds the physical layer of its Memphis cluster: "Facilities Infrastructure Engineer (Data Center Infrastructure)," "Electrical Engineer (Data Center Infrastructure)," and "Mechanical Engineer (Data Center Infrastructure)" all posted for Memphis, TN / Southaven, MS — hardware-layer expansion that stands apart from the software-and-model hiring dominant at most peers.
Google DeepMind runs three parallel tracks in one day: "Senior Mechanical Engineer, Robotics" (Mountain View), "Research Scientist, Agent Post-Training" (London), and "Threat Modeler Lead, CBRNE" (London) appeared together, reflecting simultaneous bets on physical robotics, agent training, and frontier-risk safety.

Company moves

Anthropic posted 9 new roles, the most telling being three Threat Intel Manager positions assigned to distinct threat domains: CBRN-E & advanced weapons, influence operations & surveillance, and model exploitation & fraud. The granularity — three separate managers, three separate threat ontologies — suggests an internal classification framework already mature enough to warrant dedicated headcount per category. Google DeepMind's "Threat Modeler Lead, CBRNE" appeared on the same day in the same CBRNE domain, which is unlikely to be coincidental. Anthropic also added "Partner Enablement Lead, System Integrators" and "Strategy & Operations, Applied AI - AMER," continuing to build the regional commercial layer around its Applied AI business.

Microsoft AI led volume with 29 additions and 32 closures, the highest absolute churn of the day. The spread is global and functionally mixed: solutions architects in India, a sales director in Malaysia, a strategic account technologist in Montreal, a solution architect manager covering UK & Ireland. One structural outlier is "Senior or Principal Process Design Kit (PDK) Engineer" in Redmond — PDK engineering sits at the chip-design-to-fabrication interface and is directly relevant to Microsoft's quantum computing hardware program. A "Workforce AI Solution Engineering Manager" posting in Shanghai signals continued enterprise AI deployment work in the China market.

NVIDIA added 11 roles against only 4 closures, among the cleanest net-positive ratios today. Four of those additions are explicitly AV: three "Senior Perception Engineer - Autonomous Vehicles" (all Santa Clara), plus "Senior Software Engineer, World Model Systems Engineering - Autonomous Vehicles." A "Senior AI Security Researcher" listing spanning 6 locations aligns with similar security-research postings at Anthropic and Microsoft, suggesting AI security research is becoming a standard function rather than a specialty.

OpenAI added 11 roles. On the research side, "Research Engineer / Research Scientist - Personal AGI, Personality and Model Behavior" treats model personality and behavioral customization as a named research sub-discipline — a formalization of work that was previously folded into general post-training. On the revenue side, "Revenue Accounting Manager, Ads" implies the advertising business has reached a scale that warrants dedicated financial operations headcount.

Google DeepMind posted 9 new roles. Beyond the three parallel-track roles noted above, "Product and Business Strategy Principal, Gemini App" and "Senior Data Scientist, GeminiApp" indicate sustained investment in application-layer product and analytics capability — separate from the underlying model research.

SpaceXAI added 5 roles: the three Memphis data-center infrastructure engineers, a "Technical Solutions Specialist- FTC" in Dublin (a new European customer-support location), and a "Recruiting Coordinator" in Palo Alto signaling high enough internal hiring velocity to warrant coordination headcount.

Cursor posted 4 roles oriented around industry-specific field engineering: "RVP, Field Engineering, Public Sector" and "Regional Director, Field Engineering, Healthcare" indicate Cursor is segmenting its enterprise go-to-market by vertical. A "Director, AWS Alliance" points to deepening cloud-channel partnership investment.

Meta AI added 2 roles — "Computer Vision Engineer" and "AI Research Scientist - Computer Vision" — a quiet day by count but consistent with Meta's ongoing computer-vision work across AR/VR and spatial computing.

ModelBest (面壁智能) posted the only new role from a China-based lab today: "基础模型产品经理" (foundation-model product manager), based in Beijing. All other tracked Chinese labs — Moonshot AI (月之暗面 Kimi), DeepSeek (深度求索), Zhipu AI (智谱 AI), StepFun (阶跃星辰), Meituan LongCat (美团·龙猫), and Baidu AI (百度 AI) — posted zero additions. Baidu AI closed 22 positions with no new postings, the largest single-day net reduction across all 20 companies.

Mistral AI closed 5 roles with no additions — the only Western lab with a net-negative day, despite 179 additions over the prior seven days.

What it signals

The most consequential pattern today is the concurrent emergence of structured threat-intelligence functions at multiple labs. Anthropic's three Threat Intel Manager roles are not rebranded trust-and-safety generalists — they map to specific threat domains with enough precision to warrant separate management tracks. Google DeepMind's "Threat Modeler Lead, CBRNE" appeared in the exact same domain on the same day. This cross-lab convergence on threat-intelligence methodology — borrowed from national-security and cybersecurity practice — suggests the industry is responding to external pressure or internal risk signals on a coordinated timeline. The degree of specialization (threat-intel distinct from policy distinct from product safety) signals organizational maturity in AI risk management, not an early-stage response.

NVIDIA's autonomous-vehicle concentration deserves more attention than it typically receives. The company added 594 roles over the past seven days; today four of eleven new postings were in AV, specifically in perception engineering and world-model systems. These are deep technical positions, not commercial or partnership roles. Perception engineers build the sensor-fusion and object-recognition layers; world-model engineers build the learned simulation environments that increasingly replace hand-coded rule sets. Together they describe NVIDIA filling in the DRIVE software stack at the hardest technical layers — vertical integration of AV software, not just chip sales into the AV supply chain.

The divergence between global and Chinese labs is pronounced today, though the single-day picture overstates the structural gap. The seven Chinese companies in this dataset collectively hold 1,110 open positions, and their seven-day addition figures are substantial: Baidu AI added 416, Zhipu AI 237, StepFun 150. But on a 24-hour basis, the only addition was one product manager role at ModelBest. Chinese lab hiring appears to move in batches — a burst over the week followed by near-silence on any given day — while global labs post more continuously. Whether that reflects workflow differences, approval cycles, or posting conventions, the behavioral pattern is consistent enough to watch as a structural signal rather than noise.

Microsoft's posture today illustrates what scaled AI deployment looks like in headcount terms: 29 in, 32 out, with most of the additions going to customer-facing and partner roles across a dozen countries. The actual AI-research content in Microsoft's new postings is thin; the labor is going into distributing and commercializing existing capability, not building new frontier models. That is a different growth vector than Anthropic or DeepMind, where research and safety infrastructure are still the primary hiring destinations — and it reflects where those two organizations sit in the AI product lifecycle.