AI Hiring Brief · 2026-07-23

32 frontier AI labs · 143 roles added in 24h · agihunt.info/en/jobs

Today at a glance

AGI safety becomes an engineering function: Google DeepMind and Anthropic each posted three safety-adjacent roles on the same day — DeepMind's batch covers two "Research Engineer, AGI Safety and Alignment" hires (London and US) plus a "Senior Strategic Portfolio Manager, Frontier Safety, Alignment, and Collaboration," while Anthropic added an "Evals Infrastructure Tech Lead / Manager," a "Machine Learning Infrastructure Engineer, Safeguards Research," and a "Data Engineer, Safeguards." Both labs are now staffing evaluation infrastructure and organizational management around safety, not just research headcount.
Microsoft AI's "Superintelligence Advisor" reports to the CEO: "Technical Advisor, Microsoft Superintelligence (Office of the CEO)" is an unusually titled role for a company of this scale; paired with a "Principal Software Engineering Manager - Microsoft Red Team" in Israel, "Senior Firmware Engineer" posts across four cities, and 64 net additions today, it is the single noisiest signal in today's data.
OpenAI's government track accelerates: "Head of Government Cyber Integration, OpenAI for Government" and "Integrated Marketing Manager, Government" both land in Washington DC; the same wave adds roles in Singapore, Tokyo, and São Paulo, covering three geographies in a single day.
Scale AI concentrates hiring in the DC corridor: Three of five new roles carry "Public Sector" labels and are located in Washington DC, Columbia MD, and Austin — a cluster that points to government data-training contracts entering delivery rather than pilot stage.
ByteDance PICO bets on on-device XR AI: Three new roles — "视觉算法工程师（人机交互/XR多模态方向）" (vision algorithm engineer, XR multimodal), "大模型算法工程师（压缩与轻量化方向）" (large model engineer, compression and lightweight), and "多模态感知算法工程师" (multimodal perception algorithm engineer) — point toward running capable multimodal models on XR hardware at the edge.

Company moves

OpenAI added 20 roles today (62 over seven days, 733 open). The government track is the sharpest new signal: "Head of Government Cyber Integration, OpenAI for Government" and "Integrated Marketing Manager, Government" pair cybersecurity integration with government-facing marketing in the same batch. International expansion hit Singapore, Tokyo, and São Paulo simultaneously. Consumer hardware ambitions resurface via a "Systems Integration Engineer, Build Systems | Consumer Devices" under Consumer Products. On the research side, "Research Engineer/Research Scientist, Personal AGI-Model Experience" points to personalized model behavior as a discrete research area.

Anthropic added 10 today (23 over seven days, 411 open). The Safeguards cluster — "Evals Infrastructure Tech Lead / Manager," "Machine Learning Infrastructure Engineer, Safeguards Research," and "Data Engineer, Safeguards" — turns what was primarily a research function into one requiring dedicated infrastructure and data pipelines. A "[Pipeline] Product Manager, Research (Code)" signals that code-focused research is now large enough to warrant its own product management track.

Google DeepMind (谷歌 DeepMind) added 9 today (28 over seven days, 96 open). Two "Research Engineer, AGI Safety and Alignment" posts — one in London, one across San Francisco, Mountain View, and New York — plus a "Senior Strategic Portfolio Manager, Frontier Safety, Alignment, and Collaboration" represent a focused single-day safety push. Adding organizational management to a safety function typically signals it has outgrown a flat research team. A "Research Scientist, Reasoning" in Singapore is a rare DeepMind research node outside the US-UK axis.

Microsoft AI (微软 AI) posted 64 new roles today (172 over seven days, 521 open, 66 closures). "Technical Advisor, Microsoft Superintelligence (Office of the CEO)" is seated directly under executive leadership — an unusual title at this scale. Hardware build-out continues with "Senior Firmware Engineer" across Redmond, Mountain View, Washington DC, and Austin. CoreAI expands with "Principal Product Manager - Azure Speech (CoreAI)" and "Principal Software Engineering Manager-CoreAI." "Applied scientist 2 - Fine tuning" in Bangalore reinforces India as a growing engineering hub. The 66 simultaneous closures suggest active reallocation rather than pure growth.

NVIDIA (英伟达) added 8 today (45 over seven days, 561 open). Autonomous vehicles get leadership: "Director of Engineering - Autonomous Vehicles" and "Engineering Manager, DRIVE OS Communication Infrastructure" both post today. Silicon design continues with "Senior Mask Design Engineer - Hardware" and "Senior Mixed Signal and Analog Circuit Designer." New-grad slots — "GPU Architecture Engineer - New College Grad 2026" and "RTL Power Optimization Engineer – New College Grad 2026" — replenish the chip talent pipeline.

SpaceX AI added 5 today (22 over seven days, 219 open). "Manager, Human Data Operations" spans APAC, EMEA, and Remote US/International, signaling global scaling of data operations. "Software Engineer - Evals" and "ML Infrastructure Engineer" continue core model work.

Scale AI posted 5 roles today (15 over seven days, 193 open). "Field Engineer, Public Sector" (Austin / St. Louis / DC), "Technical Program Manager, Public Sector" (Columbia, MD), and "Technical Program Manager (Gen AI Data Engine), Public Sector" (DC) together suggest the Gen AI Data Engine has entered a delivery phase with government clients. "Head of Engagement Management" under Gen AI Operations reinforces that large-scale delivery management is now a staffing priority.

ByteDance Doubao/Seed (字节跳动豆包/Seed) added 8 today (14 over seven days, 970 open) against 109 closures. In addition to the PICO XR roles above, the batch includes two "模型训练-大模型安全群组" (model training, large model safety group) posts in Beijing and Chengdu — expanding safety-training compute to a second city.

Baidu AI (百度 AI) added 3 today (20 over seven days). "具身仿真与数据工程师（J103480）" (embodied AI simulation and data engineer) lands in the ACG department — a unit previously associated with autonomous driving — suggesting Baidu is building robotics-adjacent data infrastructure in Beijing and Shanghai.

ModelBest (面壁智能) added 2 today (20 over seven days, 130 open). "Agent沙盒工程师（环境平台方向）" (Agent sandbox engineer, environment platform direction) in Shenzhen is one of the few explicit agent runtime infrastructure roles visible among Chinese labs this week.

Meta AI, Apple AIML, Perplexity, Cursor, Cohere, and Tencent Hunyuan each posted one to three roles with no concentrated cluster. Mistral AI recorded zero additions over seven days. DeepSeek (深度求索), Zhipu AI (智谱 AI), Alibaba Tongyi Qwen (阿里通义千问), MiniMax (Hailuo), 01.AI (Yi), SenseTime SenseNova, StepFun, and Meituan LongCat posted no new roles today.

What it signals

The most consistent pattern today is the industrialization of AI safety. Anthropic's 23 additions and Google DeepMind's 28 additions over seven days are both healthy rates, and the composition of today's safety-specific roles — evaluation infrastructure leads, ML infrastructure engineers, data engineers, a strategic portfolio manager — is what makes the signal meaningful. These are not more research scientists; they are the support layer that research functions acquire when they graduate into operational status. That both labs independently showed this pattern on the same day is coincidence, but the underlying dynamic — safety becoming its own engineering vertical — appears structural.

Microsoft AI's 172 additions over seven days is a pace no other tracked organization matches this week. The simultaneous presence of a CEO-adjacent superintelligence advisory role, a red team engineering manager in Israel, firmware engineers across four US locations, and CoreAI product management suggests a lab building in at least three layers at once: executive strategy, offensive security research, and hardware. The 66 closures on the same day as 64 additions point to reallocation rather than pure expansion — the net composition of what stayed (CoreAI, firmware, security, India engineering) is more informative than the addition count alone.

Government AI appears to be crossing from sales motion into service delivery. OpenAI's combination of a cyber integration lead and a government marketing manager in DC implies a contract pipeline large enough to warrant dedicated support infrastructure. Scale AI's "Technical Program Manager (Gen AI Data Engine), Public Sector" is a delivery role, not a pre-sales one — someone needs to exist to run the project. Together, these hires suggest that the first wave of government AI contracts is producing the organizational weight that sustains a second wave.

Among Chinese labs, the contrast between ByteDance and the rest is notable. ByteDance Doubao/Seed leads on total open count at 970 but closed 109 roles against 8 additions today, a ratio that points to ongoing portfolio consolidation around XR on-device AI, safety training, and cloud services. Baidu's move into embodied simulation data engineering and ModelBest's agent sandbox role are the only substantive new-direction signals from the Chinese side this week. The majority of tracked Chinese labs — DeepSeek, Zhipu AI, Alibaba Tongyi Qwen, StepFun, Moonshot AI, MiniMax, 01.AI, SenseTime — posted nothing today, a quieter pace than the global labs by a wide margin.