AI safety moved from theory to shutdowns, open models got monetised, and data-centre spending turned into balance-sheet risk. OpenAI flagged possible critical cybersecurity risk in its upcoming Astra model, which may be able to autonomously find and exploit zero-day vulnerabilities, and paused parts of internal development. Alibaba is reportedly planning to charge large commercial users of its next open-weight Qwen model through revenue sharing. Microsoft, Meta, Oracle, Amazon and Alphabet have committed around $1.09T in future lease payments for AI data centres. Anthropic is building an in-house chip design team for Claude. And Etsy cut 12% of its workforce, mostly in product and engineering. The hiring signal: cyber safety, AI governance, inference economics, infrastructure finance and chip design.
Key Takeaways
Model safety is now a cyber function: hire AI security engineers, cyber eval engineers, model safety researchers, AppSec and vulnerability specialists, AI red-team engineers and secure deployment/sandboxing engineers
Open-weight does not mean free: teams need open-model deployment engineers, model evaluation specialists, AI platform engineers, legal and commercial AI product leads, inference optimisation engineers and multi-model architecture specialists
Senior AI infrastructure roles now carry financial accountability: AI infrastructure finance, data-centre procurement, FinOps and capacity planning, power and grid strategy, GPU cluster operations and infrastructure programme management
Model companies are moving into hardware: AI chip design engineers, hardware/software co-design specialists, compiler and runtime engineers, ML systems engineers, performance and efficiency engineers and silicon programme managers
Etsy's 12% cut puts good product and engineering talent back on the market, but companies will prioritise people who connect product work to revenue, automation, trust, marketplace quality or operational efficiency
Add a 30-minute 'AI risk ownership' interview station scoring risk triage, access control thinking, cyber-safety awareness, model evaluation design, logging and auditability, stakeholder communication, and human escalation and rollback planning
This week: add AI risk ownership to senior interviews, build a Qwen/open-model commercial-risk checklist, add lease and capacity awareness to AI infra hiring, and map embodied AI and chip-inference talent
Show Notes
OpenAI flags possible critical cybersecurity risk in its upcoming Astra model, which may be able to autonomously find and exploit zero-day vulnerabilities, and pauses parts of internal development while triggering stronger safety protocols
Alibaba reportedly plans to ask large commercial users of its next open-weight model, Qwen3.8-Max, for a share of revenue generated from it, as high-performing Chinese open-weight models compete with US frontier labs
Microsoft, Meta, Oracle, Amazon and Alphabet have committed around $1.09T in future payments under leases that have not yet begun, mostly for AI data centres, and these are not yet showing as full lease liabilities on balance sheets
Anthropic confirms it is building an in-house chip design team for Claude, hiring across hardware and software to co-design chips and models while still using AWS, Google, Nvidia and AMD hardware
Etsy cuts around 12% of its workforce, roughly 220 employees, mostly in product and engineering, with at least 16 weeks of severance and healthcare support
Smaller signals: Obsidian Security $85M Series D at $1.1B, Volta Infra $2.4B valuation with a $10B European AI cloud partnership and $5B Azora initiative, AMD acquires inference chip startup Taalas, DeepSeek invests $20.8M for 2.31% of Unitree, Siemens Energy posts a record quarter on AI data-centre power demand
Quick bytes: a reported US ban on Chinese data-centre devices; Oracle's AI ratings gamble; Google shifts Demis Hassabis to chief scientist/chairman with Koray Kavukcuoglu taking operational leadership; Anthropic names Mariano-Florentino Cuéllar chief global affairs officer; Apple reportedly testing CXMT memory chips; corporate AI buyers pressed to prove returns
AI Tool Spotlight: Loxo — AI recruiting platform combining sourcing, CRM, ATS, outreach, people data and AI agents with an 800-million-person talent graph and candidate rediscovery
Frequently Asked Questions
What does the Astra pause mean for hiring?
Model safety is no longer only about hallucinations, bias or policy compliance. A model that can independently discover and exploit vulnerabilities needs a real cyber-safety function with evals, containment, monitoring, escalation and incident response.
Why does Alibaba charging for Qwen matter?
Open-weight models are becoming commercial infrastructure rather than a cheap backup. Teams need people who understand licensing, hosting, monetisation, evaluation, cost and compliance, with commercial, legal and technical sign-off on every deployment.
How do $1.09T in lease commitments affect AI infrastructure hiring?
AI infrastructure spending is a long-term financial commitment, not just capex. Senior infra specs should include cost, capacity, utilisation, reserved capacity, energy exposure and vendor lock-in ownership.
What does Anthropic's in-house chip team signal?
AI hiring is drifting toward the full stack: model, runtime, chip, memory, networking, power and cost. Expect demand for chip design, hardware/software co-design, compiler and runtime, ML systems and silicon programme roles.
How should a team pilot Loxo this week?
Pick one hard-to-fill AI security, platform or infrastructure role, define 6 must-have and 4 strong-signal criteria, build a 50-person shortlist from new sourcing and CRM rediscovery, audit the top 20 and bottom 20, and compare against a fresh LinkedIn Recruiter-only search.