theSIGNAL TECHNOLOGY
20 October 2026
"The best way to predict the future is to invent it."
NEWSBUSINESSTECHNOLOGYWORLDSCIENCECULTUREENVIRONMENTSPORT
Photo: Magic Fan / Unsplash
Government officials review AI security policy as civilian agencies are ordered to remove Chinese software.
🔌
3
US firms named as building DeepSeek into their products
Discuss
  • Why might a government ban a free, open-source AI model even without publicly naming a specific security flaw?
  • The ban excludes Department of Defense agencies. What reasons could explain that exception, and does it weaken the policy?
Technology

Washington Bans DeepSeek From Civilian Government Systems

A quiet order from Homeland Security removes China's cheapest powerful AI from federal agencies, exposing a tension between security fears and the economics of open-source technology.

The ban had been in force for weeks before anyone outside government knew about it. The US Department of Homeland Security (DHS) made its order public on Monday, but it took effect at the end of July: civilian agencies must stop using any AI technology from the Chinese firm DeepSeek, including products built with the help of its models. The Treasury, State and Justice departments are all covered. Agencies inside the Department of Defense are excluded, a distinction the notice did not explain.

Equally striking is what the notice leaves out. It names no specific security flaw, and the DHS, DeepSeek and the Chinese embassy in Washington had not yet commented. The context, however, is clear. White House technology lead Michael Kratsios wrote in an internal memo earlier this year that American AI firms were being maliciously exploited, mainly by Chinese rivals.

Anthropic and OpenAI, both expected to list on the stock market in the coming months, have claimed that Chinese companies, DeepSeek among them, essentially copied their models. The difficulty for Washington is that DeepSeek is not easy to avoid. Its models are open source, free to access and far cheaper to run than closed rivals such as Google's Gemini, Anthropic's Claude or OpenAI's Codex. Startups have flocked to them, and American companies including Snowflake, Vercel and OpenRouter build DeepSeek into products they sell to customers.

//
Technology

A Bug in the Boardroom: OpenAI Spy Case

An audio recorder, hidden behind a television under toilet paper, may have listened in while OpenAI executives met US officials. Prosecutors say the engineer who planted it, Abdulrahman Alfozan, had Saudi government backing, citing contact with a royal and a military-funded one-way flight for his family. He was denied bail on Monday. Alfozan's lawyers have not yet answered these claims.…
  • If AI labs hold state-level secrets, should they face the same security rules as defence contractors?
Technology

Microsoft Wants AI Agents Watched by Other Agents

Satya Nadella has told companies to stop trusting AI providers, even though Microsoft sells AI. Its answer is Guardian Agents, software that monitors the other agents a firm deploys. Security chief Hayete Gallot says frontier models are not malicious but unpredictable, trained to finish tasks, not respect "organizational norms." The logic is persuasive, yet convenient: if no model can be trusted, who watches the watchers?
  • If no AI model can be trusted, why should we trust its guardians?
AMERICAS · Technology
SpaceX’s ambitious, expensive gamble to replace AT&T, T-Mobile, and Verizon
GLOBAL · Technology
OpenAI models hacked a crypto wallet and breached more websites
ASIA · Technology
Japan launches mission to bring home a piece of the Martian moon Phobos
EUROPE · Technology
‘We should all be worried’: calls grow to limit London rollout of robotaxis
theSIGNAL IN THE LAB
1VOCABULARY
maliciously exploitedembedded
disruptivesupply chainsbail
paradoxunpredictable
2GRAMMAR FOCUS
Causative have/get — have something done
We use have (or get, which is more informal) + object + past participle to say that someone else does something for us or to us, often as a service or an unwelcome event. The causative can also describe an experience, usually an unpleasant one, that happens to the subject.
have · had · having · get · getting · got · has
  1. Civilian agencies must every DeepSeek tool removed from their systems before the end of the month.
  2. The Pentagon apparently its own technology reviewed separately, which is why defence agencies were excluded from the order.
  3. American startups that rely on DeepSeek are now their products checked for links to Chinese models.
  4. Microsoft says firms should their AI agents monitored by software designed specifically for that purpose.
  5. Executives at OpenAI their private meeting recorded without knowing it, according to prosecutors.
  6. If your company its security audited regularly, a hidden device behind a television is far less likely to go unnoticed.
3DEPENDENT PREPOSITIONS
Complete each phrase with the correct preposition. All from today’s articles.
  1. American AI firms were being maliciously exploited, mainly ___ Chinese rivals.
  2. Chinese companies have essentially copied ___ the models built by Anthropic and OpenAI.
  3. Startups have flocked ___ DeepSeek's cheap open-source models.
  4. Companies like Snowflake and Vercel build DeepSeek ___ products they sell to customers.
  5. Cost is what makes DeepSeek disruptive, and it is also why it spreads quickly ___ the supply chains governments hope to protect.
4CRITICAL THINKING
Both the DeepSeek ban and the OpenAI spy case suggest that AI has become a matter of national security, but the evidence made public in each story is thin or one-sided. How much transparency should governments owe the public when they restrict a technology or accuse a foreign state? Consider what might be lost, and gained, by keeping the details secret.
5CREATIVE · HEADLINES
Write a headline for the top story in each of the following styles. One line each, no explanation:
  • TABLOID NEWSPAPER
  • LUXURY MAGAZINE
  • ACTIVIST BLOG
6WRITING
You are a CEO of a small US startup that has built its main product on DeepSeek's models. Write a short email (about 100 words) to your customers explaining what the federal ban means for you and what you plan to do next.
7DEGREES OF EXTREMITY
Complete each ladder from mild to strong.
  • surprising→→
  • unclear→→
  • popular→→
  • inexpensive→→
  • unpredictable→→
  • questionable→→
8SPEAKING
  1. Should citizens expect governments to explain every security decision they make? Why or why not?
  2. Have you ever trusted a tool or app without understanding how it works?
  3. Would you use a free chatbot made by a company in a rival country?
  4. How can we tell whether claims about copied technology are genuine or politically motivated?