AI Hiring Brief · 2026-09-20

32 frontier AI labs · 127 roles added in 24h · agihunt.info/en/jobs

AI Hiring Brief · 2026-09-20

32 frontier AI labs · 127 roles added in 24h · agihunt.info/en/jobs

Today at a glance

Microsoft's mass reshuffle dwarfs everyone else: Microsoft AI added 105 roles and closed 138 in a single day (274 added over 7 days) — by far the largest volume of any lab tracked, and the new postings lean heavily toward Technical Program Manager, compliance, and security-assurance titles rather than research.
Safety governance builds out in parallel at the top two labs: OpenAI posted two Model Policy Manager roles (Multimodal Safety, National Security) in its Safety Systems org, while Anthropic added a Product Manager, Safeguards (Account Integrity & Abuse) and a Product Manager, Safe Access — both labs are staffing policy enforcement, not just research, on the same day.
Meta keeps investing in custom silicon: a new Silicon Packaging Engineer (SIPI) role under AI Infrastructure continues Meta's pattern of hardware-adjacent hiring around its own chip programs.
ByteDance's Seed team doubles down on desktop agents: 字节跳动豆包/Seed (ByteDance Doubao/Seed) posted 桌面端Computer Use Agent研发工程师/专家 (Desktop Computer Use Agent Engineer/Expert) and 桌面端Agent Sandbox研发工程师/专家 (Desktop Agent Sandbox Engineer/Expert) alongside an AI Agent前端开发工程师 (AI Agent Frontend Engineer) — a coordinated push into agentic desktop control.
Commercial and public-sector reach keeps expanding: Anthropic's Customer Success Manager, Public Sector (National Security) in Washington, DC and Perplexity's Sales Leader, Sales Development, Enterprise both point toward GTM build-out rather than model work.

Company moves

OpenAI added 2 roles (53 over 7 days) against 1 closure. Both new postings — Model Policy Manager, Multimodal Safety and Model Policy Manager, National Security — sit in Safety Systems in San Francisco, a narrow but pointed hire into policy enforcement for multimodal outputs and national-security-adjacent model use.

Anthropic added 4 roles (39 over 7 days, 2 closed): Product Manager, Safeguards (Account Integrity & Abuse); Customer Success Manager, Public Sector (National Security) in Washington, DC; Finance & Strategy Lead, Claude Platform (API); and Product Manager, Safe Access. The mix spans trust-and-safety product work and public-sector commercial support in the same day.

Google DeepMind (谷歌 DeepMind) added 1 role (26 over 7 days, 1 closed) — a Software Engineer, GenAI position in Zürich, continuing to build out its Zürich GenAI engineering presence.

Microsoft AI (微软 AI) added 105 roles and closed 138 (274 added over 7 days) — the heaviest churn of any company tracked today. New roles are dominated by Technical Program Manager, compliance, and security-assurance titles (Technical Program Manager – Strategy & Transformation, Senior Compliance Manager, Security Assurance Program Manager, Microsoft's Principal CJIS Lead), with a smaller set of applied-AI roles (Applied Scientist – CoreAI Voice Agent, Software Engineer: AI Experiences Search) mixed in — this reads as a large-scale refresh of program-management and compliance functions rather than a research push.

Meta AI added 1 role (16 over 7 days, 2 closed): Silicon Packaging Engineer (SIPI) under AI Infrastructure, keeping custom silicon packaging in its active hiring line.

NVIDIA (英伟达) added 2 roles (30 over 7 days, 2 closed): a Senior Developer Relations Lead for AI-enabled drug discovery and a PhD new-grad Research Scientist in networking research — science-adjacent and academic recruiting rather than product roles.

Apple AIML (苹果 AIML) surfaced no new roles today but closed 1, following 21 additions over the past 7 days — a pause after an active week.

SpaceXAI added 1 role (19 over 7 days, 1 closed): a Construction Scheduler for its Memphis-area Data Center group — physical data-center build-out rather than model work.

Perplexity added 3 roles (9 over 7 days, none closed): Member of Technical Staff, AI Products (early career); Consumer Support Specialist; and Sales Leader, Sales Development, Enterprise — product, support, and enterprise sales moving together.

Cohere added 1 role (9 over 7 days): Executive Producer, Executive Communications — a communications hire, not a technical one.

Scale AI surfaced no new roles today but closed 1, against 10 additions over the past 7 days.

Baidu AI (百度 AI) surfaced no new roles today but closed 2, against 18 additions over the past 7 days.

ByteDance Doubao/Seed (字节跳动豆包/Seed) added 6 roles and closed 9 (30 over 7 days) — the most active China-based lab today. New postings cluster around agents: AI Agent前端开发工程师 (AI Agent Frontend Engineer), 桌面端Computer Use Agent研发工程师/专家 (Desktop Computer Use Agent Engineer/Expert), 桌面端Agent Sandbox研发工程师/专家 (Desktop Agent Sandbox Engineer/Expert), plus core C++/Rust and frontend engineering roles — a clear build-out of desktop agent infrastructure under Seed.

MiniMax (海螺) (MiniMax Hailuo) added 1 role (2 over 7 days): Research Lead, Large Language Models, based in San Francisco — a US-based research-leadership hire, notable for a China-based lab.

The remaining 18 companies — Mistral AI, Cursor, Lovable, Together AI, Cerebras, Pika Labs, Aleph Alpha, Genmo, Moonshot AI (月之暗面 Kimi), DeepSeek (深度求索), Zhipu AI (智谱 AI), StepFun (阶跃星辰), ModelBest (面壁智能), Meituan LongCat (美团·龙猫), Tencent Hunyuan (腾讯混元), 01.AI/Yi (零一万物), SenseTime SenseNova (商汤日日新), and Alibaba Tongyi Qwen (阿里通义千问) — showed no visible hiring activity today.

What it signals

Over the past 7 days, the 19 global labs in this dataset added 537 roles combined, versus 98 across the 13 China-based labs tracked. Even stripping out Microsoft's outsized 274, global additions (263) still run well ahead of China's total — though the gap narrows once StepFun (19), Baidu (18), ModelBest (10), Moonshot (8), and 01.AI (8) are counted, showing the China-side activity is spread across several labs rather than concentrated in one.

Microsoft's numbers are the outlier of the day: 105 added and 138 closed in 24 hours is a volume no other company approaches, and the content of the new postings — TPM, compliance, security-assurance, and a "Principal CJIS Lead" role — points to organizational reshuffling around governance and program management rather than a model-building surge. It sits alongside a much smaller set of applied-AI hires (voice agents, AI-powered search), suggesting the AI product work continues but isn't what's driving the headcount churn.

The safety-and-policy roles at OpenAI and Anthropic are worth reading together: both labs are hiring policy managers and safeguard product managers in the same window they're also hiring public-sector and platform-finance staff. That pairing — safety operations scaling alongside commercial and government expansion, not behind it — is consistent with both labs treating governance as a deployment requirement rather than a research afterthought.

On the applied side, agent tooling is the clearest cross-border theme: ByteDance's Seed team is building out desktop Computer Use Agent and Agent Sandbox roles, while Perplexity and Microsoft each added at least one agent- or voice-agent-adjacent posting. Combined with Meta's continued silicon-packaging hire, the week's data points to infrastructure (custom hardware, agent runtimes) and governance (safety/compliance roles) as the two areas seeing the most concentrated hiring attention, more so than any single new product category.