THE DIRECT ANSWER

Anthropic’s September 10 threat-intelligence report examines selected activity from December 2025 through August 2026. Its categories include cyber operations, influence efforts, surveillance, fraud, weapons-related misuse and model distillation. The report concerns cases the company investigated or disrupted, rather than a representative sample of all AI use.

Read the dated primary source ↓

How to read the evidence

Provider reports can reveal patterns visible inside a particular service. They cannot, by themselves, establish the prevalence of a behaviour across the internet. Changes in detection, reporting or account enforcement can also affect what appears in a published case collection.

The defensive takeaway

For organisations deploying AI, the useful questions are about account monitoring, permission boundaries and escalation procedures. Review how unusual activity would be detected and investigated without assuming every legitimate user resembles a reported threat actor. Detailed case claims should remain attributed to the reporting company.

What this means in practice

A threat report is evidence about the cases its investigators observed, not a census of all AI activity. Its value lies in the described methods, warning signs and limits of detection. Readers comparing reports should examine scope and collection periods before treating a change in reported cases as a change in the overall prevalence of misuse.

COMMON QUESTIONS

Questions readers ask

Does this threat report measure all AI misuse?

No. It describes selected investigated activity. Its scope and collection method limit what can be inferred about wider prevalence.

SOURCES & NOTES

Follow the evidence.

Source-based coverage of the dated announcement below, with Promptwire analysis. Company-reported claims are attributed to their source. The update date reflects added context and answers; it does not change the original event date.

  1. anthropic.com — official reference

Public sources support the factual background. Analysis and illustrative exercises are identified in the text. How we assess AI · Editorial standards

Spot a factual error or an outdated detail?