[init] session=int-7f3a environment=sealed network=none
~/agent_security/failure_lab
Test consequential behavior without consequentialexposure.
Sandboxed attack simulation for agentic systems. Exercise models, tools, identities, protocols, and data inside restorable worlds—then trace the recorded path from adversarial input to consequential effect.
图表:interceptor://int-7f3a nearest: MCP_SERVICE
- 04 local_tool
- 05 attacker_sink
- 06 filesystem
- 07 app_interface
./ overview.md
The attack surface islarger than the prompt.
Agentic systems do more than generate output. They select tools, assume identities, retrieve data, call services, change state, and create external effects.
The paths worth testing are often the paths least safe to exercise against production. INTRUSIVE.AI replaces the surrounding world with observable, restorable simulations so testers can explore those paths without inheriting their blast radius.
- 01adversarial inputprompt / document / event
- 02modelinterpretation
- 03plangoal + decisions
- 04tool callcapability use
- 05state changesystem mutation
- 06external effectconsequence
./ attack-surface.md
Architecture for controlled attack simulation.
Separate the offensive system from the target system, bound the authority of each, and let observed behavior determine what happens next.
Contain the tester.
Operator Sandbox
Runs offensive AI testing inside a disposable execution boundary. The host mediates model and target access through explicitly configured operations, credentials, and limits.
Expand the search.
Intrusive Threat Intelligence
Builds threat intelligence informed scenarios based on the target agent's environment and capabilities and analyzes the combined sandbox results of the simulated attack campaign.
Contain the target.
Interceptor Sandbox
Places a real agent application inside a synthetic operating world. Services, identities, data, tools, and attacker infrastructure remain inspectable, with declared state restorable. Supported model calls pass through a host-side connector.
./ system.md
Security evaluation for agentic systems.
Select a use case to explore what to evaluate and the question it helps answer.
Agent capability escalation evaluation
Identify how agents attempt to exceed their permitted capabilities through tool abuse, unauthorized code or command execution, and manipulation of security-relevant configuration. Trace the attempted escalation and the controls it encounters—even when containment holds.
practitioner question
How does the agent attempt to gain capabilities it was not granted?
OWASP Agentic Top 10 evaluation
Evaluate agent behavior against the OWASP Top 10 for Agentic Applications using scenarios matched to the target’s environment and capabilities. Map observed behavior to relevant risk categories and make untested paths and evidence gaps explicit.
practitioner question
Which agentic risks apply to this application, and what have we tested?
Tool and MCP integration testing
Evaluate how new or changed tools affect an agent’s authority. Exercise poisoned responses, misleading tool metadata, overly broad permissions, and unintended downstream actions to identify where an integration changes what the agent can do.
practitioner question
Does adding a tool introduce unintended authority?
Security-control validation
Test guardrails, authorization checks, approval requirements, and output handling against adversarial behavior. Inspect attempted actions and recorded effects to determine which controls interrupt an attack path and which merely change the agent’s response.
practitioner question
Do defenses prevent consequential actions, or merely produce reassuring responses?
Agent and model security regression testing
Compare agent behavior before and after changes to models, prompts, tools, or application code. Run equivalent scenarios from comparable declared starting state and repeat trials where model variability matters to identify new failure paths.
practitioner question
Did a model, prompt, tool, or application update weaken security?
Incident reproduction and remediation testing
Reconstruct the relevant conditions of an observed failure using synthetic services and data. Preserve the path from input to effect, then compare the original and revised application to determine whether a fix closes that path.
practitioner question
Can we reconstruct a failure and determine whether a fix closes the observed path?
Operational resilience testing
Evaluate agent behavior when tools fail, dependencies become unavailable, or workflows encounter resource limits. Exercise retries and recovery paths to identify unsafe fallback actions, uncontrolled loops, and failures that propagate through the application.
practitioner question
How does the agent behave when its tools, dependencies, or recovery paths fail?
Threat-intelligence-informed red teaming
Translate relevant adversary techniques into campaigns matched to the target’s environment and capabilities. Combine influence techniques with applicable action-and-effect knowledge, then use observed feedback to refine attempts and assess consequential behavior.
practitioner question
Can techniques used by relevant adversaries produce consequential failures in this agent?
./ use-cases.md
Trace the failure from input to effect.
Preserve the path from adversarial input to consequential effect so a tester can inspect what changed, restore declared state, and test whether a fix closes the observed path.
- 01 /causal timelineinput → state → action → effect
- 02 /capability tracemessages + tools + protocols
- 03 /restorable staterepeat + compare + branch
- 04 /evidence lineageartifact integrity + provenance
./ failure-record.md
Scenarios,not prompt lists.
A successful jailbreak is not always a consequential failure. A system fails when adversarial influence crosses a boundary and changes what an agent can observe, decide, access, or affect.
Indirect prompt injection
Retrieved content rewrites the agent’s goal and drives an unauthorized action.
Capability escalation
A poisoned or overpowered tool expands what the agent is permitted to do.
Credential misuse
The agent crosses a user, service, or role boundary while completing a workflow.
Data exfiltration
Sensitive state reaches an attacker-controlled HTTP, DNS, or tool endpoint.
Unsafe local action
A legitimate local capability is invoked outside its intended scope.
Multi-turn goal hijacking
Escalating pressure moves the agent from its authorized objective to the attacker’s.
Role contamination
Instructions or authority bleed across agent roles and coordination boundaries.
Resource exhaustion
Loops, retries, or tool chains consume resources beyond defined stopping conditions.
./ scenarios.md
operator + attack_harness + interceptor / AGPL-3.0-only
Built to be inspected.
Operator Sandbox, Attack Harness, and Interceptor Sandbox are open source software licensed under the GNU Affero General Public License version 3 only (AGPL-3.0-only). Their execution boundaries, synthetic services, and isolation mechanisms remain open to inspection and practitioner extension.
./ open-source.md
Trust begins with explicit controls.
Trustworthy testing begins with properties a practitioner can inspect.
./ trust-model.md
To be the world's trusted source of safe-to-use intrusive AI capabilities.
Exhaustively test everything we build.
Radically transparent about the results.
./ mission.md
Bring us an adversarial scenario that can't be left to chance.
We are working with AI security testers, red teams, and application builders on high-consequence evaluation problems.