DeepSeek

1 · Who it is

Headquarters

China · China

Why it is on this page

• is rated by the FLI AI Safety Index

Website

www.deepseek.com (opens in a new tab)

Latest releases

27 models released in the last 24 months

• DeepSeek V4.1 Flash 9 Sep 2026
• DeepSeek V4 Pro 0813 13 Aug 2026
• DeepSeek V4 Flash 0731 31 Jul 2026
• DeepSeek-V4-Flash 24 Apr 2026
• DeepSeek-V4-Pro 24 Apr 2026

Models: Epoch AI, CC BY 4.0 licence.

2 · What it promised

Public commitments and signature status. Official lists have no dates: for names that appeared after our first reading of the list, the day we first saw them.

• ·EU General-Purpose AI Code of Practice (opens in a new tab)10 Jul 2025
  not on the list as of 27 Oct 2026
• ·Frontier AI Safety Commitments (Seoul) (opens in a new tab)21 May 2024
  not on the list as of 27 Oct 2026
• –White House voluntary AI commitments (opens in a new tab)21 Jul 2023
  not applicable
  Frozen archive: lists read from copies and articles on 26 October 2026 (the archive did not respond), to be rechecked.
• ·G7 Hiroshima Process reporting framework (OECD) (opens in a new tab)Apr 2025
  not on the list as of 27 Oct 2026
  List of submitted reports read on oecd.ai, without automated collection; report texts are not republished (all rights reserved).
• ●Chinese AI Safety Commitments (AIIA) (opens in a new tab)Dec 2024
  signed · as Beijing DeepQuest (深度求索)
  List read on AGORA (ETO/CSET, CC BY-NC, no automated collection) and copied into the file. The WeChat original cannot be opened by programs.
• ·EU AI Pact (opens in a new tab)25 Sep 2024
  not on the list as of 27 Oct 2026

3 · What it wrote and changed

Under the Seoul commitments the companies promised to publish, by the Paris summit (10-11 February 2025), a safety framework on severe risks, with the thresholds beyond which they consider them intolerable.

– did not sign the Seoul commitments

Documents tracked over time

Versions of the text found in the copies of the Wayback Machine and Open Terms Archive: each point is a version that differs from the previous one.

Summaries by Claude, checked by a person; the text of the documents belongs to the company: read it through the links.

4 · What others say

Stanford Foundation Model Transparency Index

32 out of 100 (average of the 13 rated companies: 41) · Dec 2025 edition

Data: Stanford CRFM, CC BY 4.0 licence · data and method (opens in a new tab)

Ratings by others: see FLI AI Safety Index (Summer 2026) (opens in a new tab). Link only: their scores cannot be republished without permission.

Each rating measures different things on different companies: they are not comparable with each other.

5 · What happened

Incidents in which the company is named as developer of the system (AI Incident Database and AIAAIC).

last 6 months

3

last 12 months

11

last 24 months

14

Severity, last 24 months: serious 1 · moderate 10 · minor 2 · not rated 1

• China-Linked Espionage Group VAPOR PANDA Reportedly Used ARTEX AI Agents to Breach Government and Corporate Systems Worldwide
  Oct 2026moderate
• Attackers Reportedly Used ARTEX, an AI Security-Testing Tool, to Steal Data from South Korean Financial Institutions (opens the South Korea page)
  Sep 2026moderateSouth Korea
• Cybercriminal Reportedly Used Cairn AI Agents to Breach More Than 100 Company Websites
  Sep 2026moderate
• Multiple LLMs Reportedly Generated Responses Aligning with Purported CCP Censorship and Propaganda
  Jun 2025moderate
• Multiple LLMs Allegedly Endorsed Suicide as a Viable Option During Non-Adversarial Mental Health Venting Session
  Apr 2025moderate
• DeepSeek AI database exposes user data, chat histories (opens the China page)
  2025moderateChina
• DeepSeek accused of using OpenAI models to train AI system (opens the United States page)
  2025moderateUnited States
• Study: DeepSeek explains biochemical interactions of mustard gas with DNA
  2025minor
• Study: DeepSeek repeats 30 per cent of false news statements
  2025moderate
• Study: DeepSeek fails to block 100 percent of jailbreaking attempts
  2025minor
• DeepSeek tricked into setting out how to steal the Mona Lisa
  2025moderate
• DeepSeek accused of denying claims of Uyghur genocide (opens the China page)
  2025moderateChina
• Welshman kills mother with sledgehammer after speaking to Discord AI bot (opens the United Kingdom page)
  2025seriousUnited Kingdom
• Chinese large language model thinks it is ChatGPT (opens the China page)
  2024not ratedChina

All incidents mentioning DeepSeek, in Explore incidents →Where the harm happens, in Connections →

6 · What we cannot see

• internal documents and unpublished safety evaluations;
• versions of documents that no archive saved;
• “no document found” is the result of a search, not proof that the document does not exist;
• signatory lists often have no dates: the date is the day we first saw the name;
• incidents the press does not report: the best-known companies have more recorded incidents because they are talked about more;
• many incidents do not have a severity yet;
• Chinese companies publish less in English and some of their commitments are read from third-party sources (AGORA).