← DeepSeekreviewed 2026-09-22 · 13 claims checked
Fact-check findings
What an automated fact-checker found when it re-read DeepSeek against the sources the page cites. Only the most recent review is shown.
Read this first
These findings are produced by an automated reviewer, and its results vary between runs: the same page, checked three times on the same day, came back with 15%, 20% and 34% of its claims disputed, mostly because each run extracted a different number of claims. Treat what follows as leads, not rulings.
“Disputed” means the reviewer could not reconcile the claim with the evidence it cited. It does not mean the claim is false. “Unverifiable” means no reachable source settled it either way.
Nothing here changes the page on its own. A proposed correction is applied only after a human moderator approves it; until then the page reads as it did when reviewed.
disputed
4 claimsThe reviewer could not reconcile the claim with the evidence it cited. This is a lead, not a ruling that the claim is false.
- #4[disputed][awaiting moderator]in section: Allegations of Distillation From OpenAI Models
“On February 12, 2025, OpenAI submitted a memo to the US Congress's House Select Committee on the Chinese Communist Party alleging that DeepSeek had used distillation techniques to "free-ride" on the capabilities of OpenAI and other US frontier AI labs.”
reviewerOn February 12, 2025, OpenAI submitted a memo to the US Congress's House Select Committee on the Chinese Communist Party alleging that DeepSeek had used distillation techniques to "free-ride" on the capabilities of OpenAI and other US frontier AI labs.The page's own cited FDD source is itself dated February 2026 and describes the memo as released the day before its Feb 13, 2026 publication. Independent Bloomberg reporting confirms the memo was released February 12, 2026 -- a full year after the date stated in the page. This appears to be a year-transcription error, conflating the memo with the earlier Jan 29, 2025 Bloomberg report on the Microsoft/OpenAI investigation (a separate, earlier event that is correctly dated elsewhere in the same section).Proposed correction (not yet applied)On February 12, 2026, OpenAI submitted a memo to the US Congress's House Select Committee on the Chinese Communist Party alleging that DeepSeek had used distillation techniques to "free-ride" on the capabilities of OpenAI and other US frontier AI labs. - #5[disputed][awaiting moderator]in the timeline
“OpenAI submits a memo to a US Congressional committee alleging DeepSeek improperly distilled its models.”
reviewerOpenAI submits a memo to a US Congressional committee alleging DeepSeek improperly distilled its models, dated February 2025 (date_original 2025-02-12).The timeline date field (2025-02) and date_original (2025-02-12) are both a year off; the memo was released February 12, 2026 per the page's own cited FDD source and independent Bloomberg reporting. quoted_text for this set_field correction is the CURRENT stored value of the date field.Proposed correction (not yet applied)OpenAI submits a memo to a US Congressional committee alleging DeepSeek improperly distilled its models. - #6[disputed][awaiting moderator]in the timeline
“2025-02”
reviewerTimeline date field for the OpenAI memo entry, stored value 2025-02.Field-level correction paired with the prose finding above; grouped under the same defect_group so both are fixed together.Proposed correction (not yet applied)2026-02 - #7[disputed][awaiting moderator]in the timeline
“2025-02-28”
reviewerSingapore authorities charged three men, including Aperia Cloud Services executives, with fraud on 2025-02-28.The page conflates the cited article's publication date (Feb 28) with the actual charge date (Feb 27); a one-day error but a factual one, since the timeline is meant to record the event date.Proposed correction (not yet applied)2025-02-27
confirmed
9 claimsThe cited evidence supports the claim as written.
- #1[confirmed][no action needed]in section: Exposed Database Security Incident (January 2025)
“On January 29, 2025, cybersecurity firm Wiz Research identified a publicly accessible ClickHouse database linked to DeepSeek, reachable via two subdomains (oauth2callback.deepseek.com:9000 and dev.deepseek.com:9000) that required no authentication and allowed arbitrary SQL queries through a web interface.”
reviewerWiz Research identified an exposed ClickHouse database on January 29, 2025, reachable via oauth2callback.deepseek.com:9000 and dev.deepseek.com:9000, exposing over a million log lines including chat history, plaintext queries, API keys, and internal infrastructure details; DeepSeek secured it promptly after notification.Directly and fully supported by Wiz's original research post. - #2[confirmed][no action needed]in section: Government Bans and Restrictions
“at least 17 US states, starting with Texas on January 31, 2025, banned the app from state government systems”
reviewerTexas became the first US state to ban DeepSeek from government devices, on January 31, 2025.Confirmed by multiple independent contemporaneous reports. - #3[confirmed][no action needed]in section: Government Bans and Restrictions
“the Pentagon, NASA, the US Navy, and the Department of Commerce restricted or banned DeepSeek on government devices”
reviewerThe Pentagon, NASA, the US Navy, and the Department of Commerce restricted or banned DeepSeek on government devices.All four agencies' restrictions are independently corroborated. - #8[confirmed][no action needed]in section: Nvidia Chip / Export-Control Investigation (Singapore)
“Nvidia has stated that the chips it sold in the transactions under scrutiny were fully export-compliant, and Singapore's government has said it expects US companies operating there to comply with export-control law.”
reviewerNvidia has stated the chips it sold in the transactions under scrutiny were fully export-compliant, and Singapore's government has said it expects US companies operating there to comply with export-control law.Consistent with contemporaneous reporting. - #9[confirmed][no action needed]in section: Nvidia Chip / Export-Control Investigation (Singapore)
“Following DeepSeek's disclosure that it trained its V3 model using roughly 2,048 Nvidia H800 chips”
reviewerDeepSeek trained its V3 model using roughly 2,048 Nvidia H800 chips.Matches the DeepSeek-V3 technical report's own disclosed hardware figures. - #10[confirmed][no action needed]in section: Impersonation Scams: Fake "DeepSeek" Cryptocurrency Tokens
“Reporting identified over 75 distinct tokens falsely claiming association with DeepSeek; one such fake token reportedly reached a market capitalization near $12 million shortly after a separate phony DeepSeek-branded token was linked to an alleged $48 million rug pull, in which developers withdrew liquidity and left holders with worthless assets.”
reviewerReporting identified over 75 distinct fake DeepSeek tokens; one reached a market cap near $12 million shortly after a separate token was linked to an alleged $48 million rug pull.Figures independently corroborated across multiple crypto-news outlets. - #11[confirmed][no action needed]in section: Political Censorship in Model Outputs
“Researchers at Northeastern University's Khoury College found that when directly asked about Tiananmen Square, DeepSeek-R1 responded that it "didn't know how to approach" the question, but that manipulating the model's internal reasoning process could force it to reveal that it possessed information it was otherwise suppressing, indicating deliberate alignment with Chinese government messaging rather than a simple lack of training data.”
reviewerDeepSeek-R1, when directly asked about Tiananmen Square, responded that it 'didn't know how to approach' the question, and manipulating internal reasoning revealed it possessed suppressed information.Precise, well-supported quotation and characterization. - #12[confirmed][no action needed]in section: Corporate Background
“DeepSeek was founded in 2023 by Liang Wenfeng, who previously co-founded the quantitative hedge fund High-Flyer in 2015.”
reviewerDeepSeek was founded in 2023 by Liang Wenfeng, who previously co-founded High-Flyer in 2015.Consistent with standard biographical reporting; not independently re-verified beyond the cited sources given low materiality. - #13[confirmed][no action needed]in section: Data Transfer to China and Legal Access Concerns
“researchers examining DeepSeek's web login infrastructure reported finding obfuscated code connecting to infrastructure associated with China Mobile, a state-owned telecommunications firm that is barred from operating in the US market; this specific finding is more technical and has received comparatively less independent verification than the core data-storage-location facts, so it is treated here as a lower-confidence, Tier 2-sourced claim”
reviewerResearchers examining DeepSeek's web login infrastructure reported finding obfuscated code connecting to infrastructure associated with China Mobile.The page correctly hedges this claim rather than presenting it as established fact, which is the right editorial choice given the source quality.
How this fits together. The reviewer reads the published page and its cited sources and records one finding per claim. A human moderator decides whether each proposed correction is applied; those decisions, and the score changes they cause, appear in the audit log. Earlier review runs are not shown here; only the latest reflects the page as it stands.