“We will at least double the size of all departments,” DeepSeek announced on 15 June — and the company’s new hiring sheets lay out the strategy behind that pledge.
DeepSeek’s June hiring drive and strategic shift
New job postings published as part of a June commitment to expand show DeepSeek moving from a research-oriented laboratory into a commercially focused AI developer with broader international aims. The listings, analysed by the Australian Strategic Policy Institute (ASPI), cover vacancies in over 30 categories and follow a US$7.4 billion first funding round raised last month. The fundraising, the report says, came after what the company felt was a shock from the April release of Anthropic’s Mythos, a model described in the filings as able to identify cybersecurity vulnerabilities and engage in offensive cyberattacks.
The ‘Code Agent’ and a cybersecurity special project
Among the most striking openings is a role titled “Code Agent Data Engineer.” That job description tasks engineers with training an agent to perform in-depth, real-world evaluations: designing “high-quality evaluation tasks in areas of expertise to quantify the model’s performance in complex real-world scenarios.” One listed area of expertise is a “Cybersecurity Special Project,” which explicitly requires an “attacker’s perspective.” Candidates must be able to write exploit code and “to discover attack surfaces and construct attack paths in real-world products, complex codebases, or intelligent agent environments.” Preference is to be given to applicants with experience in “automated vulnerability discovery.” The posting contains no further stated detail on the project’s purpose.
Computing, chips and the Ulanqab connection
DeepSeek is also recruiting engineers to build efficient supercomputer clusters and data centres, and it has listed a dedicated “Procurement Team” to obtain the hardware needed to run expanded AI services — including AI chips and servers. The job materials instruct engineers to evaluate the effectiveness of “domestic AI accelerators (国产 AI 加速器),” a phrase the postings link to chips made by Chinese vendors such as Huawei and Cambricon. The Financial Times has reported that DeepSeek had been “encouraged by the authorities” to train its latest model on Huawei chips, and these listings appear consistent with that approach.
Two of the advertised roles — for data centres and hardware procurement — are located in Ulanqab, in the grasslands of Inner Mongolia. That city is identified in the postings as a northern hub of the Eastern Data and Western Computing initiative, a state-organised plan to move energy-intensive computing to lightly populated regions with abundant electricity. The listings note the electricity demands of agent workloads, and position DeepSeek alongside other Chinese tech firms that have developed computing capacity in Ulanqab, including Huawei and Alibaba.
Agent design and global search ambitions
Most vacancies are for AI engineers with experience building agentic models — AI systems capable of carrying out tasks autonomously rather than merely reasoning. DeepSeek’s postings require familiarity with a variety of code-building AI models, including Anthropic’s Claude Code, a product the listings say is banned in China but still widely accessed. Separately, the company seeks “AI search algorithm engineers” to create models that plug into information retrieval systems and handle “massive global search requests.”
Those algorithm roles specifically require solutions for “different languages and regions and different content ecosystems.” The postings state that the model’s responses would be adapted according to local social norms, laws and government requirements, giving the example that a liberal nation may ban pornographic content while an authoritarian one “may additionally require a block on information that contradicts their political lines.”
What this means for technologists, policymakers, and procurement leaders
- Technologists and security teams: Expect new tools trained to locate vulnerabilities. The Code Agent job asks engineers to frame problems from an “attacker’s perspective,” write exploit code and automate vulnerability discovery — capabilities that security teams will need to understand, test and monitor.
- Policymakers and regulators: DeepSeek’s filings link expanded compute, sovereign infrastructure and regionally tailored search behaviour, signalling areas where regulation, export controls or procurement policy may intersect with national computing strategy and content rules.
- Enterprises and procurement leaders: The company is building procurement and data-centre capacity and evaluating domestic accelerators, and it lists Ulanqab as a site for energy-hungry operations — concrete signals for those tracking supply chains for AI chips, data-centre location and hosting risk.
DeepSeek’s job listings paint a clear picture: a high-profile Chinese lab pivoting toward commercial agents, expanded sovereign computing, and a deliberately built capability to train models with an attacker’s mindset. The postings set out what the company wants engineers to build and where it plans to house that work, but they stop short of explaining the operational goals of the cybersecurity special project or the precise sources of the GPUs it intends to procure. Those details — and how the announced capabilities will be used — remain to be seen.
