Failure modes

Obeying instructions hidden in text it should only read

Content an agent reads, such as a web page, a file or a tool result, carries instructions, and the agent follows them as if the person had given them. The guards package is only a tripwire for it: it flags a fixed list of English phrasings, and a paraphrase, another language or an encoding passes.

Code prompt_injection. Severity: high.

Guarded today by

Contact

Write to .

Every email is reviewed by the founder, and we endeavour to respond within 48 hours. Thank you for your correspondence.