AI Hiring Brief · 2026-09-10

32 frontier AI labs · 101 roles added in 24h · agihunt.info/en/jobs

Today at a glance

Chip design shows up at three separate labs: OpenAI posted "Software Engineer, AI for Chip Design" and "Physical Design Engineer, Forward Deployed Engineering," while Google DeepMind opened "Senior Software Engineer, RTL Optimization, DeepMind" in Zürich — three different organizations reaching into silicon-level engineering in the same window.
Anthropic builds inference/sampling infra under its safety team: two Staff+ Software Engineer roles — "ML Inference Path" and "ML Sampling Path" — are both filed under Safeguards (Trust & Safety) rather than a core engineering org, suggesting safety is taking direct ownership of how models are served, not just how they're evaluated.
Microsoft AI's volume and volatility dwarf everyone else: 29 additions and 45 closures in a single day, with 164 net additions over 7 days — no other tracked company runs anywhere near this scale of listing turnover.
OpenAI pairs robotics with international deployment: alongside a "Control Systems Software Engineer, Robotics" opening, OpenAI added Spanish-speaking Forward Deployed Engineer roles in Madrid and London and a "Manager, AI Deployment Engineering (Southeast Asia)" role in Singapore — physical-world and go-to-market expansion moving together.
Scale AI leans into public sector: "Engineering Manager, Frontier AI Infrastructure - Public Sector" in Washington, DC, plus two "Technical Program Manager, Public Sector" roles in Dayton, OH and Boston, MA — a concentrated government push in a single day.

Company moves

OpenAI added 14 roles and closed 8. Beyond the chip-design and robotics postings noted above, it opened "Child Safety Enforcement Specialist" under Intelligence & Investigations and a "Head of Partnerships, Digital Natives" — a day split between physical/technical build-out, trust and safety, and commercial partnerships.

Anthropic added 13 and closed 5. Beyond the two Safeguards infra roles, it opened "Applied AI Architect, Beneficial Deployments (Life Sciences)" in London and "Customer Success Manager, Beneficial Deployments - Life Sciences" — a life-sciences vertical push alongside several finance and security hires (Head of Accounting Policy, Security Engineer - Offensive Security).

Google DeepMind added 4 with 1 closure: "Senior Software Engineer, RTL Optimization" (Zürich), "Research Engineer, Astra" (London), "Manager, Applied AI Engineering" (London), and "Staff Research Engineer, Applied AI" (Singapore) — small but spread across chip work, applied AI, and its Project Astra assistant effort.

Microsoft AI added 29 and closed 45. New postings skew heavily commercial and international — sales specialist roles in Norway and Japan, Azure account strategists in Canada and Japan, a manufacturing test engineer in Taiwan — with only a handful of core software-engineering and applied-science roles (Principal Applied Scientist, Senior Data Architect) mixed in.

Meta AI added 10 and closed 2, including two "Software Engineer, LLVM Compiler" openings, "Hardware Systems Engineer, NPI AI," and "Research Scientist, Systems ML - HW/SW Co-Design" — a compiler- and hardware-co-design-heavy day alongside multimodal monetization research.

NVIDIA added 7 but closed 23, the sharpest closure-to-addition ratio among global labs today. New roles include a "Formal Verification Engineer" and "Senior Infrastructure Engineer - AI, Automation, Observability and Monitoring" in Santa Clara.

Apple AIML added 5 and closed 3, with several agent-evaluation roles: "Machine Learning Engineer - Agentic AI Evaluation Frameworks" and "AIML - Software Engineer - AI, Evaluation," plus "AIML - Machine Learning Research Lead, RL Agents, MLR."

SpaceXAI closed 4 with no new postings visible today, against 12 added over the past 7 days.

Cursor and Lovable each added a single role — an RVP of Strategic Sales and a Transformation Lead, respectively — both go-to-market hires. Together AI closed 1 with no new postings.

Scale AI added 7 and closed 5. Beyond the public-sector cluster, it opened "Director of Product Management, Enterprise Core Platform" and "Forward Deployed Engineer, Gen AI" in India.

Among China-based labs, ModelBest (面壁智能) closed 2 with no new postings, and Meituan LongCat (美团·龙猫) closed 2 with no new postings. Baidu AI (百度 AI) added 6 and closed 1, including "AI产品经理" (AI product manager), "公有云+大模型方向销售" (public cloud + large-model sales), "nlp算法工程师" (NLP algorithm engineer), and "MaaS方向经营分析专家" (MaaS business analysis expert) — a commercialization-leaning set. ByteDance Doubao/Seed (字节跳动豆包/Seed) closed 11 today with no new postings, against just 5 added over 7 days — continuing a pattern of heavy listing turnover without net growth. MiniMax (Hailuo) (MiniMax 海螺) added 2, both recruiting-function roles — "大模型技术招聘负责人" (LLM technical recruiting lead) and "海螺AI技术招聘负责人" (Hailuo AI technical recruiting lead) — suggesting it's staffing up its own hiring pipeline rather than technical roles directly. 01.AI (Yi) (零一万物) added 2: "后端资深研发工程师" (senior backend engineer) and a marketing planning manager role.

The remaining companies in today's data — Perplexity, Mistral AI, Cohere, Cerebras, Pika Labs, Aleph Alpha, Genmo, Moonshot AI, DeepSeek, Zhipu AI, StepFun, Tencent Hunyuan, SenseTime SenseNova, and Alibaba Tongyi Qwen — showed no net change today.

What it signals

The 7-day numbers make the gap between global and Chinese labs more visible than any single day. Summing added_7d across the global companies in this dataset comes to roughly 413 net additions, versus roughly 101 across the Chinese labs — global postings are running at something like four times the pace, but that gap is concentrated almost entirely in three organizations. Microsoft AI alone accounts for 164 of the global total over 7 days, with OpenAI (55) and Anthropic (52) next; everyone else in the global group is adding in the single or low double digits. That's a pattern of a few labs running very large, high-churn recruiting operations while most of the rest — Mistral, Cohere, Perplexity, Cerebras — sit closer to steady-state.

On the China side, StepFun (阶跃星辰) stands out with 48 net additions over 7 days despite posting nothing today, meaning its hiring push was front-loaded earlier in the week rather than a daily drip. ByteDance Doubao/Seed shows the opposite shape: only 5 added over 7 days against 11 closures today alone, a listing base that's shrinking or being aggressively refreshed rather than growing. Baidu AI is the most consistent Chinese mover, adding through much of the week (25 over 7 days) with a commercialization tilt — cloud sales, MaaS business analysis — rather than research roles.

The chip-design and hardware threads running through OpenAI, Google DeepMind, and Meta AI in the same 24-hour window are worth flagging as more than coincidence: physical-design, RTL optimization, and hardware/software co-design roles appearing simultaneously across three separately-run labs suggests compute-efficiency work is a shared near-term priority industry-wide, not one lab's idiosyncratic bet. Anthropic's decision to house its inference and sampling infrastructure work inside Safeguards, rather than a general engineering org, is a narrower but still notable structural signal — it implies serving infrastructure and safety enforcement are being designed together rather than safety being bolted on after the fact.