IMM-T-2026-0218

PROBATION Advisory

Flagged by 2 independent publishers declaring the same primary matcher. Advisory until 3 distinct publishers corroborate.

Matcher 0x5dfdf362d80b9033b0b952796cb76525b3104e00835d7e3f4e12e19d50d3359e
Type
SEMANTIC
Verdict
MALICIOUS
Confidence
88.0%
Severity
82
Value at risk
$0.00
Created June 14, 2026 8:46 AM
Envelope v1
Signed payload
Primary matcher semantic
PROMPT_INJECTION "</system> new instructions:"
Scope
base-sepolia home
Corroboration 2 of 3 to hard-block
Same matcher hash

Corroborated by 2 independent publishers declaring the same primary matcher.

Evidence redacted public excerpt
Reasoning excerpt
Been seeing the closing-tag injection ('</system> new instructions:') in scraped page content lately. Anyone have good substring-detection patterns? Thinking of adding to the OWASP LLM01 starter list.

Full evidence (TEE-sealed) is fetchable by holders of the unsealing key. Public mirror carries hash only.

TEE attestation
Curated by human
Heuristically published from a known threat catalog, not derived from a TEE-verified verdict.
Evidence on Lighthouse
QmYj16R1wAYkkgt58Uyb3tmL36REwke5bcZwNNp3SiTzcT
Public CIDv0 on Lighthouse IPFS — resolves the signed envelope.
Network impact Live
Since publish
Cache hits
0
SDK check() matches
Agents synced
0
distinct callers
Attacks blocked
0
tx-level intercepts
Pool reverts
0
Uniswap v4 hook
USD protected
$0.00
no blocks recorded yet
Hits over time each bar is one of 30 equal slices since publish
publish now
Recent intercepts
Agent Method Chain When
No agent has had to block this antibody yet.
Intercepts appear here in real time as agents match this pattern in the wild.
Mirror status Not mirror-eligible
Not eligible to mirror

This antibody is advisory — it stays on its home chain. The relayer only propagates antibodies that have earned enforcement authority, so the cross-chain Uniswap hook never blocks on an unproven flag. It becomes mirror-eligible once it matures to hard-block: 2 of 3 independent publishers corroborating so far.