DHS Bans DeepSeek Across US Agencies as of Oct. 19 [2026]
![DHS Bans DeepSeek Across US Agencies as of Oct. 19 [2026]](https://trendintech.com/wp-content/uploads/2026/10/dhs-bans-deepseek-federal-agencies-2026-photo-3.webp)
The Department of Homeland Security has ordered every civilian federal agency to strip DeepSeek products and any third-party tools built on DeepSeek models out of their information technology systems. The directive, posted in a DHS notice on October 19, 2026, also blocks agencies from signing new contracts that rely on the Chinese AI startup’s software, closing off a company that has spent the past two years undercutting US rivals on price.
DHS Secretary Markwayne Mullin signed the underlying order on July 24, 2026, according to the department’s own notice. Monday’s posting is the public rollout of that order, and it marks the first time Washington has moved to erase a specific Chinese AI model from the plumbing of the federal government rather than simply warning agencies to be careful. Engadget was among the first outlets to report the ban on DeepSeek’s use in federal civilian agencies, and the story quickly spread across financial and policy press as the practical mechanics came into focus.
For an agency ecosystem that increasingly experiments with large language models for everything from document triage to automated decision support, the order forces an immediate inventory question: does any system, vendor platform, or contractor tool touch a DeepSeek model, directly or indirectly. The answer determines whether an agency is now out of compliance.
Don't miss new tech stories on Google
Add TrendinTech once in the Google app and our stories appear in your news suggestions.
What the DHS Directive Actually Orders
Stripped of legal language, the DHS order does three things. First, it tells civilian federal agencies to remove DeepSeek products from their information technology systems. Second, it extends that removal requirement to third-party products and services that use DeepSeek models under the hood, even if the end product does not carry the DeepSeek name. Third, it bars civilian agencies from acquiring DeepSeek products or covered third-party services going forward, which effectively closes DeepSeek out of new federal government contracts.
That middle piece, the third-party clause, is the part procurement officers and federal contractors are likely to find most disruptive. A growing number of software vendors that sell into government quietly route some workloads through DeepSeek’s open-weight models because they are cheap to run and easy to self-host. The DHS order does not carve out an exception for vendors who simply integrated DeepSeek as a backend rather than selling it as a branded product. Cyberscoop’s reporting on the notice, published as DHS moves to ban DeepSeek from federal civilian agencies, framed the order as reaching well past DeepSeek’s own consumer app and into the supply chain of tools built on top of it.
What the order does not do, based on everything DHS and the outlets covering it have published so far, is apply beyond civilian agencies. The notice addresses civilian federal agency systems. It does not establish, in the public record so far, that the removal requirement extends to the Department of Defense, intelligence agencies, state and local governments, or private companies that are not federal contractors. Anyone extending this to “the US government has banned DeepSeek outright” is reading past what has actually been confirmed.
The Paper Trail: A July Order, an October Notice
One detail that has gotten lost in some of the faster headlines is the gap between when the order was signed and when it became public. Mullin signed the directive on July 24, 2026. The notice describing it did not appear until nearly three months later, on October 19, 2026. That lag is normal for federal rulemaking, where an agency head signs an internal directive and then works through the notice-and-process steps before it is published for the public and for contractors to see. Bloomberg Law’s coverage, which ran under the headline US Moves to Bar China’s DeepSeek From Federal Agency Systems, captured both dates and treated the October posting as the operative moment agencies and vendors need to react to.
Below is a timeline of the publicly confirmed dates tied to this action, pulled directly from DHS’s own notice and the outlets that first reported it.
| Date | Development | Source |
|---|---|---|
| May 7, 2025 | Senators Bill Cassidy and Jacky Rosen introduce the Protection Against Foreign Adversarial Artificial Intelligence Act, aimed at restricting federal use of DeepSeek and similar tools | U.S. Senate |
| July 24, 2026 | DHS Secretary Markwayne Mullin signs the order directing removal of DeepSeek products from civilian agency IT systems | DHS |
| October 19, 2026 | DHS notice describing the order is posted publicly, triggering compliance obligations and press coverage | DHS |
| October 19 to 20, 2026 | Bloomberg, Engadget, Cyberscoop, Firstpost, and other outlets report the order internationally | Multiple outlets |
Notice how much of the legislative groundwork predates the actual DHS action by more than a year. The Cassidy-Rosen bill, introduced in May 2025 as S.1638, proposed a broader statutory prohibition rather than an agency-level procurement order. Its text, available through Congress’s own records, states plainly that “no Federal contractor with an active Federal contract may use the DeepSeek application or any successor application or service developed or provided by High Flyer or any entity owned by High Flyer” for covered federal work. You can read the full bill text on Congress.gov. DHS’s July order and October notice are narrower than that bill in scope, since they cover civilian agency systems and contracts rather than every federal contractor regardless of agency, but they move in the same direction and suggest the legislative push had real traction inside the executive branch even before lawmakers finished voting on it.
S.1638, Protection Against Foreign Adversarial Artificial Intelligence Act of 2025
(a) Prohibition.--Except as provided in subsection (b), no Federal
contractor with an active Federal contract may use the DeepSeek
application or any successor application or service developed or
provided by High-Flyer or any entity owned by High-Flyer, for the
fulfillment, assistance, execution, or...
Why Homeland Security Is Worried About DeepSeek
DHS grounded the order in two overlapping concerns: that sensitive federal data could be transferred to entities tied to the People’s Republic of China, and that DeepSeek’s software could carry malware or other security vulnerabilities. Neither concern is unique to DeepSeek among Chinese-origin technology products, but DeepSeek’s rapid adoption inside the United States, driven by models that are free to download and cheap to run compared with offerings from American labs, gave the concern more urgency than it might have carried for a less widely used product.
The “Multiple Threat Vectors” Problem
DHS described DeepSeek’s models as posing “multiple threat vectors,” a phrase that appeared in the department’s own notice and was picked up by Firstpost in its report on the DeepSeek removal order over data security concerns. That language is deliberately broad. It covers data exfiltration risk, where prompts or documents fed into a DeepSeek-hosted model could in theory be accessible to parties in China. It covers supply chain risk, where a compromised or backdoored model weight file could sit inside federal infrastructure for months before detection. And it covers dependency risk, the general discomfort inside DHS and allied agencies with having any part of the federal government’s AI stack rely on a company subject to Chinese law and potential state direction.
None of the public reporting so far has disclosed a specific incident, breach, or forensic finding tied to a DeepSeek deployment inside a federal agency. The order reads as a precautionary, forward-looking procurement decision rather than a response to a confirmed compromise. That distinction matters for how seriously the private sector should treat the underlying technical risk versus how seriously it should treat the political and compliance risk, which is immediate regardless of whether a specific incident ever surfaces.
What’s Covered, and What Isn’t
Because the order has already generated some confusion in early social media reaction, it is worth laying out exactly what the public record supports and what it does not.
| Scope question | Status |
|---|---|
| Civilian federal agency IT systems | Confirmed: DeepSeek products must be removed |
| Third-party products and services built on DeepSeek models | Confirmed: Covered by the removal and acquisition bar |
| New civilian agency contracts involving DeepSeek | Confirmed: Barred going forward |
| Department of Defense and intelligence community systems | Not established in public reporting |
| State and local government systems | Not established in public reporting |
| Private companies with no federal contract | Not covered by this order |
| Specific removal deadline or dollar figure | Not disclosed publicly |
| Permanent statutory ban versus agency procurement order | Currently an agency-level directive, not confirmed as permanent law |
That last row is worth sitting with. An agency procurement and systems-removal order can, in principle, be modified or rescinded by a future DHS leadership team without an act of Congress. The Cassidy-Rosen bill would lock a version of this restriction into statute, which is a meaningfully different and harder-to-reverse outcome. Whether lawmakers finish that process is one of the open questions this order does not answer on its own.
How DeepSeek Got Here: A Fast Rise Out of Hangzhou
DeepSeek built its reputation by releasing open-weight language models that matched or approached the performance of closed, proprietary systems from American labs at a fraction of the training and inference cost. That pricing advantage is precisely why the company shows up anywhere cost-conscious buyers, including federal procurement offices and the contractors that serve them, go looking for an AI backend. The company’s most recent public release, a search-and-reasoning oriented update covered in detail in TrendInTech’s report on DeepSeek’s newest flash model and open API, highlighted that the company has kept shipping competitive, cheaply priced models even as political scrutiny of its US footprint intensified.
That is the tension at the center of this story. DeepSeek’s commercial appeal and its political liability are the same thing: a Chinese company offering Western buyers dramatically lower prices than domestic competitors can match, which is exactly the kind of dependency Washington has spent the better part of a decade trying to avoid in strategic technology sectors.
The Bigger Picture: Washington’s Broader AI Trade Standoff With Beijing
The DeepSeek order lands in the middle of a much wider set of trade and technology restrictions between the two countries. The US has spent much of 2026 tightening export controls on chips and related equipment, a push that has touched everything from advanced semiconductors to the solar supply chain, as detailed in coverage of China’s own retaliatory posture around export controls on dozens of tech categories. On the chip side, American AI accelerator makers have simultaneously watched their share of the Chinese market erode as domestic alternatives from Chinese chipmakers gain ground, a dynamic examined in reporting on how Nvidia’s position in China has shifted as Beijing pushes its own suppliers.
The DeepSeek order is the software and model-layer counterpart to those hardware-layer restrictions. Where export controls try to slow China’s access to the chips needed to train frontier models, the DHS order tries to slow the inverse flow: a finished, cheap, highly capable Chinese model making its way deep into American government infrastructure. Both policies rest on the same underlying premise, that strategic dependency on the other country’s AI stack, in either direction, is something policymakers in Washington now treat as a national security variable rather than a purely commercial one.
AI Vendor Status With the Federal Government
To put DeepSeek’s new status in context, it helps to see where other major AI labs stand relative to federal procurement rules as of this order.
| AI lab or model provider | Headquarters | Federal civilian agency status after this order |
|---|---|---|
| DeepSeek | Hangzhou, China | Barred: removal required, new contracts blocked |
| OpenAI | San Francisco, United States | Not affected by this order |
| Anthropic | San Francisco, United States | Not affected by this order |
| Google DeepMind | Mountain View, United States | Not affected by this order |
| Mistral AI | Paris, France | Not affected by this order |
That table is less about picking winners than about illustrating how narrowly the order is drawn. It names one company, DeepSeek, and the entities it controls, such as High-Flyer. It is not a general rule about open-weight models, models trained outside the United States, or AI companies with foreign ownership in general. A European open-weight lab is not swept up in this order simply by virtue of releasing open weights. The restriction is tied specifically to DeepSeek’s PRC-based ownership and operations.
Reaction From the AI Policy World
The order arrives amid an ongoing argument inside the US AI industry about whether open-weight models, regardless of country of origin, carry inherently different risks than closed, hosted systems. That argument has been loudest among American labs with the most to gain from steering buyers toward proprietary platforms. One executive at a major American AI lab has already gone on record warning that open-sourced AI model weights carry risks that closed systems do not, a position detailed in reporting on warnings about open-source AI risk, and the DeepSeek order gives that argument a concrete, government-sanctioned example to point to.
At the same time, the broader legislative push for AI safety rules in Washington has continued on a separate but related track, including a proposed federal AI safety framework pitched to lawmakers that would apply across the industry rather than targeting a single foreign company, covered in detail in reporting on a proposed AI safety bill pitched to Congress. The DeepSeek order and that kind of general safety legislation are aimed at different problems, foreign dependency versus domestic model behavior, but both reflect the same underlying mood in Washington: that AI policy has moved from a hands-off posture to active, specific intervention in how federal agencies and private industry choose and deploy models.
Market Impact: A New Compliance Burden for Federal Contractors
For federal systems integrators and software vendors, the immediate effect of the order is an audit requirement, not a technology swap. Every contractor with civilian agency work now needs to document whether any component of its stack, including any subcontractor’s stack, calls a DeepSeek model at any point. Because DeepSeek’s models are open-weight and inexpensive to self-host, they have been quietly embedded inside smaller software products in ways that are not always obvious from a vendor’s marketing materials. Finding those dependencies will take time and will likely surface integrations that program managers did not know existed.
Who Has to Act, and By When
Public reporting has not disclosed a specific removal deadline, which leaves agencies and contractors without a hard date to plan against. In practice, federal procurement orders of this kind typically get implemented through follow-on guidance from the agency’s chief information officer and through amendments to existing task orders and contract vehicles, a process that can take weeks to months even after the underlying policy is final. Contractors should expect contracting officers to start asking about DeepSeek dependencies in routine compliance reviews well before any formal deadline is published.
Vendors that build AI features for government customers using a mix of models, routing some queries to cheaper open-weight systems and others to premium hosted models, now have a strong incentive to drop DeepSeek from that mix entirely for any government-facing product line, even if the removal order technically applies only to civilian agencies today. Nobody wants to be the contractor explaining to an agency inspector general why a DeepSeek call was still buried three layers deep in a subcontractor’s pipeline.
Historical Context: This Is Not Washington’s First Chinese Tech Ban
DeepSeek’s removal from federal systems follows a pattern Washington has applied to other Chinese technology products over the past several years, restricting networking equipment, security software, and social media platforms with ties to China on similar national security grounds. What distinguishes the DeepSeek case is the underlying asset: not hardware sitting in a data center or an app running on a phone, but model weights and the software built around them, which are far easier to copy, embed, and redistribute than a physical device. A banned router can be physically removed and verified as gone. A banned model that has already been downloaded, fine-tuned, and wrapped inside a dozen third-party tools is much harder to fully excise from an IT environment, which is exactly why the order’s third-party products clause is drawn so broadly.
This also is not the first time a government body has flagged concerns about a specific AI model before it ever reached full deployment. Earlier in 2026, UK authorities raised early concerns about a different frontier model ahead of its public release, a case covered in reporting on how the UK AI Security Institute flagged risks before a model’s launch. Taken together with the DHS order, these episodes point to a shift in how governments on both sides of the Atlantic are treating frontier AI: as infrastructure that warrants pre-deployment security review, not just post-hoc regulation after something goes wrong.
Competitive Comparison: What DeepSeek Loses, and What It Keeps
It is worth being precise about the scale of what DeepSeek is actually losing here. The US civilian federal government is a meaningful customer segment for enterprise software broadly, but DeepSeek’s core business has never been built around direct federal sales. Its growth has come overwhelmingly from developers, startups, and enterprises worldwide who adopted its open-weight models because they were free to download and cheap to run on commodity hardware. The DHS order closes one specific, prestige-heavy customer segment and sends a signal that ripples through risk-averse enterprise buyers who watch what the federal government does before making their own procurement calls. It does not touch DeepSeek’s developer community, its API business outside the United States, or its ability to keep releasing new open-weight models.
The more durable damage is reputational. Every enterprise security team now has a concrete, documented example to cite the next time someone on a product team proposes routing sensitive data through a DeepSeek-hosted endpoint. That kind of reference point tends to outlast the narrow legal scope of the order that created it.
What DeepSeek and High-Flyer Have Said
As of this writing, neither DeepSeek nor its parent company High-Flyer has issued a public statement responding to the DHS order. That silence is itself notable given the size of the US market and the directness of the security allegations DHS raised, but it should not be read as agreement or concession. Chinese technology companies facing US restrictions have historically taken days to weeks to issue formal responses, often through state-affiliated media rather than direct corporate statements, and there is no indication yet of which path DeepSeek will take this time.
Predictions: Where This Goes From Here
- Expect follow-on guidance from individual agency chief information officers within the next several weeks, translating the DHS order into specific contract language and audit checklists for existing task orders.
- The Cassidy-Rosen bill, or a successor version of it, is likely to gain renewed momentum in Congress now that an executive branch agency has effectively validated its premise through independent action.
- Other civilian agencies and potentially state governments will face pressure to adopt parallel restrictions, even without a legal requirement to do so, simply to avoid being the outlier still running Chinese-model-dependent software.
- Federal systems integrators will begin marketing “DeepSeek-free” or “sovereign AI stack” compliance audits as a billable service, mirroring how contractors previously built businesses around Huawei and Kaspersky removal compliance.
- DeepSeek’s commercial strategy outside the United States, particularly in price-sensitive markets across Asia, Africa, and Latin America, is unlikely to change, since the US federal government was never the center of its growth story.
Why This Matters Beyond Washington
The practical reach of a civilian agency procurement order is narrower than the headlines suggest, but the signal it sends is not. Private companies, universities, and foreign governments that watch US federal procurement decisions as a proxy for security best practice now have a specific, named data point: the US government decided a leading Chinese AI model was too risky to keep inside its own IT systems. That kind of decision travels well beyond the agencies it directly binds, shaping vendor selection conversations in boardrooms that have nothing to do with federal contracting at all.
It also raises the stakes for how the next major Chinese AI release gets received in Western markets. Straits Times, in its coverage of the order, noted that the move fits a broader pattern of US scrutiny toward Chinese-origin AI tools generally, not DeepSeek in isolation. Any Chinese lab hoping to sell into Western enterprise or government markets will now have to answer, proactively, questions about data handling and ownership structure that American and European competitors simply do not face in the same way.
Frequently Asked Questions
What exactly did DHS order regarding DeepSeek?
DHS ordered civilian federal agencies to remove DeepSeek products from their information technology systems, extended that removal requirement to third-party products and services that use DeepSeek models, and barred agencies from acquiring DeepSeek products or covered third-party services going forward.
When does the order take effect?
The underlying order was signed by DHS Secretary Markwayne Mullin on July 24, 2026. The public notice describing it was posted on October 19, 2026. No specific removal deadline has been disclosed publicly as of this writing.
Does this ban cover the entire US government?
No. Public reporting establishes that the order applies to civilian federal agencies. It does not establish that the Department of Defense, intelligence agencies, state and local governments, or private companies without federal contracts are covered.
Why did DHS single out DeepSeek specifically?
DHS cited the potential transfer of sensitive data to entities in the People’s Republic of China and the possibility of malware or other security vulnerabilities, describing DeepSeek’s models as posing “multiple threat vectors.”
Has DeepSeek responded to the order?
No public statement from DeepSeek or its parent company High-Flyer has been reported as of this writing.
Is this the first US government action targeting DeepSeek?
No. Senators introduced the Protection Against Foreign Adversarial Artificial Intelligence Act in May 2025, proposing a broader statutory prohibition on federal contractor use of DeepSeek. DHS’s order is the first executive branch action of this kind to be formally posted.
What happens to contractors that already use DeepSeek-based tools?
They are now out of compliance for civilian agency work and need to audit their stacks, including subcontractor dependencies, to identify and remove any DeepSeek-based components, even if those components are embedded inside third-party software rather than sold under the DeepSeek name.
Could this become permanent law rather than an agency order?
Possibly. The Cassidy-Rosen bill, still pending in Congress, would write a version of this restriction into statute, which would be harder to reverse than an agency-level procurement order. Whether that bill advances is not yet confirmed.
Related Coverage
Daniel Okafor
Daniel Okafor is the Senior AI Reporter at TrendinTech, where he covers large language models, machine learning research and the practical use of artificial intelligence across business and government. He has a particular interest in evaluation: how models are benchmarked, where those benchmarks fail and what that means for the companies betting on them.
All stories by Daniel Okafor (381)![OpenCode AI: Local Coding Agent in 14 Steps, 90 Min [2026]](https://trendintech.com/wp-content/uploads/2026/10/opencode-ai-local-coding-agent-setup-tutorial-2026-photo-3-640x360.webp)

