FraudBench is a new benchmark that tests policy‑grounded banking conversational agents against adaptive fraud scenarios. It uses a dual‑control framework and a 698‑document internal policy corpus, presenting 150 adversarial scenarios (107 public, 43 held‑out) that require agents to manage mutable account state and tool access while preventing identity, authorization, and trust manipulation. Preliminary results on four agents show attack‑security rates between 49% and 65%, highlighting weaknesses in money‑mule and first‑party fraud detection.
arXiv:2606. 18356v1 Announce Type: cross Abstract: Tool-using language-model agents introduce security failures that go beyond unsafe text: they can disclose protected objects, write persistent memory, send messages, modify databases, or trigger harmful code and tool effects.
arXiv:2606. 17555v1 Announce Type: cross Abstract: Banks simultaneously face signature-based fraud (card-not-present attacks, account takeover, ATM cloning) and behavioural financial crime (structuring, layering, mule networks, business email compromise) -- two threat families with fundamentally different detection requirements.
arXiv:2608. 09476v1 Announce Type: cross Abstract: Cowork agents may complete benign tasks while disclosing protected data, manipulating unauthorized state, invocate unauthorized API.
The paper audits self‑evolving financial agents—SkillOpt, Agent Workflow Memory (AWM), and ReasoningBank—by evaluating their performance, security drift, and execution‑interface mismatches in simulated e‑banking scenarios. It shows that while utility improves after evolution, exposure to malicious content and unauthorized state changes also rise, and that AWM’s text‑action envelope can disrupt tool execution, highlighting the need for comprehensive auditing beyond accuracy metrics.
arXiv:2605.12015v3 Announce Type: replace-cross Abstract: Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files...
The paper introduces PACE (Policy‑Attested Contract Execution), a framework that sits between large‑language‑model (LLM) based autonomous AI agents and on‑chain DeFi operations. PACE defines typed transaction intents, a deterministic policy verifier, and signed Policy Decision Records (PDRs) that cryptographically bind an approved intent, policy, and simulation report to the exact on‑chain execution bytes, providing replay and expiration protection. In evaluations across 40 tasks and six baselines, PACE achieves zero unsafe executions and zero false positives, outperforming unguarded agents by a large margin.
arXiv:2606. 30755v1 Announce Type: cross Abstract: Claw-like AI agents (e.
ClawSentry is an open‑source, framework‑agnostic security supervision gateway designed to protect autonomous large language model (LLM) agents from progressive risks that can arise at four points in the agent control loop: skill admission, invocation‑time intent, execution‑time effect, and post‑action consequence. It introduces a multi‑tier decision engine—deterministic L1, rule‑anchored L2, and read‑only L3—alongside a First‑Use Skill Package Review (FSPR) and an Agent Harness Protocol (AHP) that applies a single policy across multiple agent runtimes without modifying their internals. Evaluation on SkillInject and the SkillsSafety benchmark shows that ClawSentry significantly reduces contextual adversarial skill risk (ASR) while maintaining high task success rates (TSR).
arXiv:2608.24551v1 Announce Type: cross Abstract: Machine learning models are widely used in financial fraud and credit-risk detection, yet their adversarial robustness remains difficult to evaluate...
arXiv:2607. 08066v1 Announce Type: new Abstract: Chain-of-thought (CoT) monitoring is a promising safety mechanism for AI agents, based on the premise that visible reasoning traces can surface misaligned or deceptive behavior.
HarnessRisk is a lifecycle-oriented benchmark for evaluating safety in agent harnesses that manage tools, extensions, state, permissions, and external actions. It defines six operational phases—Harness Configuration, Capability Extension, Runtime Operation, State Persistence, Action Control, and Incident Recovery—and includes 128 sandboxed cases pairing benign user objectives with adversarial instructions. Across three harnesses, six language models, and 14 configurations, attack success rates vary from 12.6% to 80.9%, with the most vulnerable phase being Harness Configuration. "whyItMatters":"The benchmark demonstrates that safety failures can arise in multiple harness responsibilities and that even explicit risk detection does not guarantee safe action, underscoring the need for comprehensive evaluation across model and harness configurations."
arXiv:2606. 02240v1 Announce Type: cross Abstract: Indirect prompt injection in tool-use agents is a concrete production threat: LLM agents read from integrations (third-party services such as Gmail, Salesforce, or Jira accessed through tool calls) whose response content the user neither writes nor controls.