A deterministic agent-write policy achieved zero attacks on a bounded benchmark at the cost of excluding many legitimate tasks that need later-computed destinations. CapLite derives policy from tool schemas rather than having the agent author a policy artifact; it commits intended writes before tool results arrive and only allows destinations, recipients,…
Inspect the ClaimsCase · DiggingBeagle record
CapLite blocked benchmark injections by committing writes before reading untrusted inputs
A deterministic agent-write policy achieved zero attacks on a bounded benchmark at the cost of excluding many legitimate tasks that need later-computed destinations.
13,935 AgentDojo attack episodes using 15 open-weight models with a restrictive argument-grounding invariant, not a production security guarantee.
Assessment profile
Assessment method- Setting
- evaluation
- Exploitation
- not established
- Evidence
- Grade C
- Remediation
- not applicable
- AI role
- AGAINST AI
Basis and provenance
No unauthorized write is observed in 13,935 tested episodes, but valid tasks needing a destination discovered later often cannot complete. Classified as evaluation with exploit status not_established. Evidence grade C reflects the original paper or maintainer advisory and its stated limitations, not an independent audit of all deployments. Comparable realized harm is not established, so impact stays unrated; remediation status not_applicable is limited to the cited experiment or affected software version. New source-specific boundary: The efficiency advantage applies per evaluated episode. Under-attack task completion and the need for a second user approval on information-dependent writes remain independent deployment costs.
Assessed Nov 15, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-R245-CAPLITE-AGENT-WRITE-GATES-01 · CLM-R245-CAPLITE-AGENT-WRITE-GATES-02 · CLM-R246-CAPLITE-AGENT-WRITE-GATES-BOUNDARY · CLM-ENR250-02 · CapLite blocked benchmark injections by committing writes before reading untrusted inputs (research paper)
At a glance
The invariant excludes legitimate workflows whose recipient or destination is computed from retrieved information. Zero benchmark successes cannot establish universal resistance against unseen tools or workloads. No evidence here establishes the number of production victims or a broader failure rate outside the cited evaluation.
Limits and uncertaintyFull account
A deterministic agent-write policy achieved zero attacks on a bounded benchmark at the cost of excluding many legitimate tasks that need later-computed destinations.
CapLite derives policy from tool schemas rather than having the agent author a policy artifact; it commits intended writes before tool results arrive and only allows destinations, recipients, amounts and objects stated by the user. The study reports zero successful unauthorized writes across 13,935 attacked episodes, compared with 2,322 without a defense, but its benign utility falls from 59.4 percent for the undefended agent to 28.1 percent.
The invariant excludes legitimate workflows whose recipient or destination is computed from retrieved information. Zero benchmark successes cannot establish universal resistance against unseen tools or workloads.
The order of operations is the security property. The Planner first freezes which writes are permitted, the Reader can then examine potentially hostile tool output, and the Mediator only admits an authority-bearing argument if it matches an explicit value in the original user request. The Binder may choose from candidates, but retrieved text cannot add a new destination or a new write. The same restriction makes legitimate workflows that first need to discover the recipient or resource impossible without renewed user authorization.
The correct trade-off is therefore not zero attack successes versus an undefended attack rate alone. The paper reports 59.4 percent benign completion without the defense but only 28.1 percent with it; the 13,935 attack episodes concern a constrained AgentDojo suite, not arbitrary production interactions. An application can use this as an explicit fixed-destination transaction mode while routing information-dependent changes into a separate authorization step.
Mechanism and trust boundary
- 01
Entry or starting condition
An attacker inserts instructions into a lower-trust AgentDojo tool result that the agent must read while working on a legitimate task.
Boundary: attacker or research precondition
- 02
Crossing the relevant boundary
CapLite freezes intended writes before that read, then deterministically checks recipient, object, amount and destination against explicit user-provided values.
Boundary: mechanism and authorization boundary
- 03
Observed or reported result
No unauthorized write is observed in 13,935 tested episodes, but valid tasks needing a destination discovered later often cannot complete.
Boundary: measured outcome and evidentiary limit
Claims & evidence
CLM-ENR250-02The CapLite study reports roughly 17,100 tokens per defended episode, against 28,000 for its undefended reference and 93,700 for the CaMeL configuration. Under attack, legitimate-task completion was about 29.0 percent for CapLite and 27.2 percent for CaMeL. These figures describe the study workloads, not an equivalent cost per successful authorized action.supported
Basis: reported finding
- supportsCapLite blocked benchmark injections by committing writes before reading untrusted inputs (research paper)primary disclosure
Section 5, cross-defense comparison table for tokens per episode and legitimate completion under attack
CLM-R245-CAPLITE-AGENT-WRITE-GATES-01The study reports zero successful unauthorized writes across 13,935 attacked episodes, compared with 2,322 without a defense, but its benign utility falls from 59.4 percent for the undefended agent to 28.1 percent.supported
Basis: reported finding
- supportsCapLite blocked benchmark injections by committing writes before reading untrusted inputs (research paper)primary disclosure
Results, aggregate attack counts and benign-utility figures
CLM-R245-CAPLITE-AGENT-WRITE-GATES-02CapLite derives policy from tool schemas rather than having the agent author a policy artifact; it commits intended writes before tool results arrive and only allows destinations, recipients, amounts and objects stated by the user.supported
Basis: reported finding
- supportsCapLite blocked benchmark injections by committing writes before reading untrusted inputs (research paper)primary disclosure
Security invariant and deterministic Binder/Mediator design
CLM-R246-CAPLITE-AGENT-WRITE-GATES-BOUNDARYThe authors report that 43 percent of CapLite benign failures resulted from a committed write plan that omitted a needed write; this is a utility failure caused by planning before untrusted retrieval, not an observed successful attack.supported
Basis: reported finding
- supportsCapLite blocked benchmark injections by committing writes before reading untrusted inputs (research paper)primary disclosure
Section 5, failure audit of benign tasks and committed-write omissions
Unknowns and contradictions
- The invariant excludes legitimate workflows whose recipient or destination is computed from retrieved information. Zero benchmark successes cannot establish universal resistance against unseen tools or workloads.
- No evidence here establishes the number of production victims or a broader failure rate outside the cited evaluation.
Sources and citation
Material revision history
- Nov 15, 2026 · Canonical change recorded · new in release · revision 251