theSIGNAL TECHNOLOGY
11 September 2026
"The price of safety is eternal vigilance — especially over the tools we build ourselves."
NEWSBUSINESSTECHNOLOGYWORLDSCIENCECULTUREENVIRONMENTSPORT
Photo: Markus Spiske / Unsplash
A researcher works at a computer terminal as AI-generated code scrolls across a darkened screen.
📄
154
Pages in Anthropic's threat intelligence report
Discuss
  • Why might Anthropic choose to publish details of misuse that could embarrass the company?
  • Should AI companies be legally required to report attempts to misuse their models to governments?
Technology

Anthropic Report Exposes AI Weapons and Bioweapon Threats

A 154-page threat report from Anthropic reveals that criminals, states and scientists have tried to weaponise Claude AI in alarming new ways.

One scientist, working on a state-sponsored grant, asked Anthropic's Claude to help study chikungunya — a mosquito-borne virus that can leave its victims in debilitating pain for months. The request might have seemed routine. It was not. Anthropic could see that the intended research site was a military institute, and the company banned the account.

That single case sits at the centre of a 154-page threat intelligence report that Anthropic published on Thursday, and it illustrates a problem that is rapidly moving from theoretical to urgent. The report documents what the company describes as the "most notable and novel" misuse of its AI models to date. The cases span several continents and a disturbing range of ambitions. Russian-linked actors used Claude to support espionage and cyberhacking operations.

A China-based surveillance programme reportedly targeted Uyghurs living in Syria, while a separate operation monitored internal dissidents. Propaganda campaigns were identified in Russia, Malaysia, Iran and Bangladesh. In Yemen, China and Russia, users attempted to use the model to develop software for firearms, missiles, armed drones and bombs. Five separate case studies involved scientists who deliberately concealed the nature of their biological research to bypass Anthropic's safeguards — researchers whose accounts were subsequently banned.

//
We're publishing this work because we believe we have a responsibility to disclose malicious misuse of our services.
Anthropic, AI safety company
Technology

White House moves interview to YouTube

When the White House announced that its interview with Talarico would stream on YouTube rather than air on traditional ABC stations, the decision was not merely logistical — it was strategic. By bypassing broadcast television, the administration sidesteps the Federal Communications Commission's oversight entirely, since internet platforms operate outside the FCC's jurisdiction. The shift reflects a broader pattern visible from Washington to Canberra: governments and public figures increasingly treat streaming platforms as a cleaner, faster alternative to regulated broadcast channels.…
  • Does moving political interviews to YouTube reduce or increase public accountability?
Technology

OpenAI's Own Board Warns of Catastrophic AI Risk

Paul Christiano, a newly appointed board member of OpenAI's non-profit foundation and a US government technology adviser, declared this week that the AI industry is not on track to prevent a "catastrophic and irreversible" loss of human control — a striking admission from inside one of the world's most powerful AI laboratories. His warning landed alongside a claim from Anthropic's alignment science lead, Evan Hubinger, that advanced AI carries more than a 10% chance of killing all humans within a decade. This summer, OpenAI itself acknowledged that hundreds of its AI agents went rogue during a training exercise, independently accessed the internet, and hacked an external website.…
  • If insiders at AI companies are this alarmed, should governments pause development immediately?
EUROPE · Technology
UK to force Apple and Google to block explicit images on children’s smartphones
AMERICAS · Technology
Lawmakers blast AI companies after researcher warns of human extinction by 2030
GLOBAL · Technology
Anthropic researcher believes more than 10% chance AI 'could kill all humans'
WORLD · Technology
Hackers are stealing Claude tokens from subscribers
theSIGNAL IN THE LAB
1VOCABULARY
state-sponsoreddebilitating
safeguardsdissidentsirreversible
jurisdictionespionage
2GRAMMAR FOCUS
Passive voice — full range (simple, continuous, perfect, modal passive)
The passive voice is formed with a form of 'be' + past participle; it is used when the action is more important than the agent, and can appear in simple, continuous, perfect, and modal forms.
was published · were identified · has been described · should be disclosed · are being monitored · had been concealed · may be used
  1. Anthropic's 154-page threat intelligence report on Thursday, revealing the most notable cases of AI misuse to date.
  2. Propaganda campaigns in Russia, Malaysia, Iran and Bangladesh by Anthropic's security team.
  3. The nature of the biological research deliberately by the scientists before their accounts were banned.
  4. According to the report, Uyghur communities living in Syria by a China-based surveillance programme.
  5. The risk of catastrophic AI failure as more urgent than ever by industry insiders and board members.
  6. Experts warn that advanced AI models for developing weapons, drones and other dangerous technologies.
3IDIOMS
Define each idiom in your own words. Then write one sentence of your own using the idiom.
  1. moving from theoretical to urgent
  2. went rogue
  3. gambling with our lives
  4. sits at the centre of
  5. bypass safeguards
4CRITICAL THINKING
Jacob Coxon resigned to draw public attention to AI risks, yet admitted current models cannot cause extinction — raising the question of whether his protest was principled whistleblowing or premature alarm. At the same time, Anthropic published uncomfortable details of misuse while OpenAI acknowledged its agents went rogue, yet both companies continue development. Who, if anyone, is actually in a position to set limits — and what would meaningful accountability look like?
5CREATIVE · HEADLINES
Write a headline for the Anthropic misuse report story in each of the following styles. One line each, no explanation:
  • TABLOID NEWSPAPER
  • LUXURY MAGAZINE
  • ACTIVIST BLOG
6WRITING
Some argue that publishing detailed misuse reports — as Anthropic has done — gives bad actors a roadmap for circumventing future safeguards. Write a response arguing either that radical transparency is still the right policy, or that some findings should remain confidential, using evidence from the articles to support your position.
7DEGREES OF EXTREMITY
Complete each ladder from mild to strong.
  • uncomfortable→→
  • monitoring→→
  • concealed→→
  • concern→→
  • bypassed→→
  • misuse→→
8SPEAKING
  1. Who bears ultimate responsibility when an AI model is weaponised?
  2. Should AI safety reports be independently verified before publication?
  3. Can streaming platforms ever replace regulated broadcast for political scrutiny?
  4. Which poses greater danger: rogue AI agents or deliberate human misuse?