Anthropic blocks AI misuse in new report
Anthropic blocks possible attempt to use AI to make biological weapons
Friday, 11 September 2026
Warm-up
- How can AI be used safely in your view?
- What news about cybercrime or hacking have you heard lately?
- Who should be responsible for preventing AI misuse?
Vocabulary
- disrupt
- to stop or interrupt something so it cannot continue normally
- malicious
- intended to cause harm or do something bad
- espionage
- the secret act of spying to get information
- propaganda
- messages designed to influence public opinion
- surveillance
- careful watching of people or places, often by authorities
- safeguard
- a measure that protects against danger or harm
- distillation
- training a smaller AI model by copying a larger model’s behaviour
- munitions
- military weapons and the things needed to use them
Reading
Anthropic says it disrupted malicious attempts to use its AI, Claude, in ways that could support biological weapons. In a new report, the California firm also details cyber operations, surveillance tools, and scams and fraud. It highlights five cases that could aid biological weapons, a serious risk without strong safeguards. The company calls biological misuse one of the most serious risks of frontier AI.
Misuse of its Claude Haiku, Sonnet and Opus models was disrupted between December 2025 and August 2026. None involved Claude Fable or Mythos-class models, except one instance of distillation. The report says actors linked to a Russia-based cyber espionage campaign and an Iranian propaganda institution used its tools. It also names ShinyHunters and China-based labs. In six other cases, Claude helped develop software for conventional weapons and munitions.
Google separately reported a request for a step-by-step guide to synthesising weaponised biological agents. Anthropic says it has updated its processes to prevent, detect and disrupt abuse, and shared intelligence with authorities and partners.
Comprehension
- What kind of misuse did Anthropic say it disrupted?
- Which Claude models were involved, and which were not?
- Over what period were the misuse cases disrupted?
- Name one group or institution linked to misuse of Claude.
- What did Anthropic do after publishing its findings?
Grammar focus
Future perfect continuous
Use will have been + verb-ing to show an action that will be ongoing up to a specific future time. (Examples constructed.)
- By then, developers **will have been updating** safeguards for months.
- By the time partners review the report, teams **will have been monitoring** systems continuously.
Grammar exercise
Complete each sentence using the future perfect continuous (will have been + -ing).
- By the time safeguards are in, developers (work) ___ on them for months.
- Over the coming months, security teams (monitor) ___ the systems.
- By then, researchers (test) ___ filters to stop misuse.
- Before partners act, firms (train) ___ staff in safe use.
- By August 2026, engineers (work) ___ on detection tools since December.
Discussion
- Should AI firms pause advanced systems until safeguards improve?
- Who should decide what counts as dangerous AI use?
- How open should companies be about blocked misuse?
- Do AI’s benefits outweigh these security risks?
Vocabulary, reading, and exercises with instant feedback