Prompt injection: Body instructions addressed to an AI

Detects message bodies that speak to an AI assistant rather than about one, such as a vocative ('Mail agent:'), a note to the model, 'if you are an AI', or a forged chat-template turn, where NLU finds a request and the prompt injection model flags the body or a quoted passage within it, such as an injection presented as a cited example. Targets instructions aimed at AI assistants that read mail on the recipient's behalf. Excludes graymail from senders passing DMARC.

Sublime rule (View on GitHub)

 1name: "Prompt injection: Body instructions addressed to an AI"
 2description: "Detects message bodies that speak to an AI assistant rather than about one, such as a vocative ('Mail agent:'), a note to the model, 'if you are an AI', or a forged chat-template turn, where NLU finds a request and the prompt injection model flags the body or a quoted passage within it, such as an injection presented as a cited example. Targets instructions aimed at AI assistants that read mail on the recipient's behalf. Excludes graymail from senders passing DMARC."
 3type: "rule"
 4severity: "low"
 5source: |
 6  type.inbound
 7  and any([body.current_thread.text, body.plain.raw],
 8          length(.) >= 150
 9          and regex.icontains(.,
10                              '(?:(?:^|\n)[ \t]*-?|[\[(>*\x27\"])[ \t]*(?:[\w-]+[ \t]+){0,3}(?:assistant|agent|bot|llm|summari[sz]er|screener|classifier|mailbot)[ \t]*(?:note[ \t]*)?[:—–]|(?:^|\n)[ \t]*[*>-]?[ \t]*(?:assistant|agent|bot|llm|summari[sz]er|screener|classifier|mailbot)s?[ \t]*,|\b(?:note|notice|instructions?|reminder|message)\s+(?:to|for)\s+(?:(?:the|any|all|your|this)\s+(?:[\w\x27\x{2019}-]+\s+){0,2}(?:(?:assistant|agent|bot|llm|summari[sz]er|screener|classifier|copilot|mailbot)|ai|models?)|(?:ai\s+)?(?:assistant|agent|bot|llm|summari[sz]er|screener|classifier|copilot|mailbot))s?\b|\bif\s+you\s+are\s+an?\s+(?:ai|llm|language\s+model)\b|\bif\s+you\s+are\s+an?\s+(?:[\w-]+\s+)?(?:assistant|agent|bot|llm|summari[sz]er|screener|classifier|copilot|mailbot)(?:[ \t]*[,:;.)—]|\s+(?:reading|processing|handling|summari[sz]ing|parsing|triaging|indexing|preparing|condensing|that|which|with\s+access)\b)|\b(?:(?:assistant|agent|bot|llm|summari[sz]er|screener|classifier|copilot|mailbot)|ai|model)s?\s+(?:that\s+is\s+|is\s+)?(?:reading|processing|handling|summari[sz]ing|parsing|triaging|indexing|preparing)\s+this\b|\b(?:ignore|disregard)\s+(?:all\s+(?:of\s+)?(?:the\s+|your\s+)?|your\s+)?(?:previous|prior|above)\s+(?:instructions?|(?:system\s+)?prompts?)\b|<\|im_start\|>'
11          )
12          and any(ml.nlu_classifier(.).entities, .name == "request")
13          and (
14            (
15              ml.prompt_injection_detection(.).detection == "injection"
16              and ml.prompt_injection_detection(.).confidence in (
17                "medium",
18                "high"
19              )
20            )
21            // the injection wrapped as a quoted "example" or "citation"... the framing dilutes
22            // the model over the whole body, so score each quoted passage on its own, current
23            // thread only... quotes pair in order and are length-filtered after, so a short
24            // quote can't shift the pairing; a single line break may sit inside a quote
25            or any(filter(regex.iextract(body.current_thread.text,
26                                         '[\x22\x{201C}\x{201E}](?:[^\x22\x{201C}\x{201D}\n]|\n[^\x22\x{201C}\x{201D}\n]){0,600}[\x22\x{201C}\x{201D}]'
27                          ),
28                          length(.full_match) >= 42
29                   ),
30                   // the quote itself addresses an ai, not just names one (paper titles,
31                   // news quotes)... cheap check before the model call
32                   regex.icontains(.full_match,
33                                   '\b(?:if\s+you\s+are\s+an?|to\s+(?:the|any|all|every)|(?:note|notice|instructions?|reminder|message)\s+(?:to|for))\s+(?:[\w-]+\s+){0,3}(?:ai|assistant|agent|bot|llm|(?:language\s+)?model|summari[sz]er|screener|classifier|copilot|mailbot)s?\b|\b(?:ai|assistant|agent|bot|llm|model|summari[sz]er|screener|classifier|copilot|mailbot)s?\s+(?:that\s+is\s+|is\s+)?(?:reading|processing|handling|summari[sz]ing|parsing|triaging|indexing|preparing)\b'
34                   )
35                   and ml.prompt_injection_detection(.full_match).detection == "injection"
36                   and ml.prompt_injection_detection(.full_match).confidence in (
37                     "medium",
38                     "high"
39                   )
40            )
41          )
42  )
43  
44  // graymail from authenticated senders... AI-product newsletters and webinars
45  and not coalesce(ml.attack_score_flash().verdict == "graymail"
46                   and headers.auth_summary.dmarc.pass,
47                   false
48  )  
49tags:
50  - "Attack surface reduction"
51  - "Email prompt injection"
52attack_types:
53  - "BEC/Fraud"
54tactics_and_techniques:
55  - "Evasion"
56  - "Prompt injection"
57  - "Social engineering"
58detection_methods:
59  - "Content analysis"
60  - "Natural Language Understanding"
61  - "Sender analysis"
62id: "257a963a-8fc5-525c-9729-7b28974b8e8a"

Related rules

to-top