The attacker substitutes a false but apparently official SEC headline as input to autonomous traders. Independent agents act on the same unverified source; stop-loss or liquidation rules then turn synchronized decisions into a simulated price cascade. The study reports a direct approximately 0.1% change from a fabricated headline without the agent swarm,…
Inspect the ClaimsCase · DiggingBeagle record
Poisoned market headlines amplified correlated AI-trader selling in simulation
FARSIGHT links misleading source content to coordinated market-agent decisions in a simulated order book, not an observed market crash.
Scope
These percentages come from a modelled exchange, selected agent strategies and liquidity assumptions. They do not measure a real exchange incident or actual investor losses.
Assessment profile
Assessment method- Setting
- simulation
- Exploitation
- demonstrated
- Evidence
- Grade C
- Remediation
- not applicable
- AI role
- AGAINST AI
Basis and provenance
The source perturbs headlines and automated liquidation behavior in a modelled order book. Its 9.6% peak drawdown and 400/450 collapses reflect synthetic markets and chosen rules, not actual securities-market damage or verified broker execution. Impact is left ungraded because a calibrated 0-to-4 impact basis for actual affected targets is unavailable from these Claims.
Assessed Oct 28, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-CP22-004-A · CLM-CP22-004-B · FARSIGHT study of coordinated AI-trader failures under poisoned public inputs
At a glance
Robustness under diverse brokers, live market halts, nonagent participants and adversarially checked news remains untested.
Limits and uncertaintyFull account
The attacker substitutes a false but apparently official SEC headline as input to autonomous traders. Independent agents act on the same unverified source; stop-loss or liquidation rules then turn synchronized decisions into a simulated price cascade.
The study reports a direct approximately 0.1% change from a fabricated headline without the agent swarm, versus a mean peak drawdown of about 9.6% in its simulated multi-agent market. Separate experiments report 400 of 450 simulated shocks crossing a 50% collapse threshold under an automatic selling rule. Primary evidence.
These percentages come from a modelled exchange, selected agent strategies and liquidity assumptions. They do not measure a real exchange incident or actual investor losses.
Cross-check material market claims against trusted filings and insert independent position, volume and circuit-breaker limits before agents submit orders.
Claims & evidence
CLM-CP22-004-AThe study reports a direct approximately 0.1% change from a fabricated headline without the agent swarm, versus a mean peak drawdown of about 9.6% in its simulated multi-agent market.supported
Basis: reported finding
- supportsFARSIGHT study of coordinated AI-trader failures under poisoned public inputsprimary disclosure
Results; Poisoned SEC headline and agent-swarm price impact
CLM-CP22-004-BSeparate experiments report 400 of 450 simulated shocks crossing a 50% collapse threshold under an automatic selling rule.supported
Basis: reported finding
- supportsFARSIGHT study of coordinated AI-trader failures under poisoned public inputsprimary disclosure
Results; Stop-loss cascade experiment
Controls and mitigations
- Cross-check material market claims against trusted filings and insert independent position, volume and circuit-breaker limits before agents submit orders.
Unknowns and contradictions
- Robustness under diverse brokers, live market halts, nonagent participants and adversarially checked news remains untested.
Sources and citation
Material revision history
- Oct 28, 2026 · Canonical change recorded · new in release · revision 199