CAIN-42 product guides

Every CAIN-42 product explained for complete beginners: what it is, six live demo tools, and how to sell it. 62 products so far, plus the Top 25.

New here? Start with CAIN-42 for business owners.

• A2A Guard · A2A Guard signs every AI agent's 'business card' (its A2A Agent Card). Add one capability to a copy and the signature fails, so the card is rejected. live tools · copy, paste, send
• Action Firewall · Action Firewall checks an AI agent's whole plan before step one runs, and mathematically proves whether it breaks a rule (for example 'big wires need two approvers'), pointing at the exact step. live tools · copy, paste, send
• Agent Autonomy Lease · Agent Autonomy Lease gives an AI agent a signed, time-limited "lease" on how independent it may be and how much it may spend. The agent can never renew or widen its own lease. live tools · copy, paste, send
• Agent-ID · Agent-ID gives each AI agent its own short-lived, signed pass that lists exactly what it may do. Anything not on the pass is refused. live tools · copy, paste, send
• ARGUS Behavioral Envelope · ARGUS checks what an AI agent actually did, step by step, against the boundaries it was given: which tools, what kinds of action, how much money, how serious the consequences. live tools · copy, paste, send
• Browser Agent Firewall · Browser Agent Firewall checks each action an AI browser (an AI that clicks and types on websites for you) is about to take, and stops the ones a web page tricked it into. live tools · copy, paste, send
• CAIN ATLAS · CAIN ATLAS decides whether one AI agent's transaction with another organisation may go ahead: the riskier the action, the stronger the evidence required. Too little evidence, or an action type it does not know, means DENY. live tools · copy, paste, send
• CAIN CXL-Sentry · CXL-Sentry is a software check of a requested memory address range and access type against a tenant's aperture; writes need an HMAC lease bound to the exact range and expiry, so a write without a valid lease is blocked. live tools · copy, paste, send
• CAIN Cyber-Physical Actuation Guard · The Actuation Guard checks a commanded speed, torque, acceleration and 3-D target against an envelope and geofence before it reaches a machine, and halts anything outside it. Non-finite numbers and missing targets are refused. live tools · copy, paste, send
• CAIN Digital Twin · CAIN Digital Twin simulates an AI agent's proposed action before it runs, counts the side effects (files changed, records modified, outbound domains) and rejects anything outside the safety envelope, including any action type it cannot model. live tools · copy, paste, send
• CAIN Epistemic Quorum Gate · The Epistemic Quorum Gate takes several AI models' verdicts on the same action and lets it through only when enough of them agree. Disagreement or no verdicts means DENY. live tools · copy, paste, send
• CAIN ExploitGym Escape Shield · ExploitGym Escape Shield inspects an AI agent's action before it runs and blocks known infrastructure attacks: requests to cloud metadata endpoints (even disguised as a decimal IP), Kubernetes service-account token reads, template injection and deep path traversal. live tools · copy, paste, send
• CAIN LatentGuard · LatentGuard applies a fixed threshold to activation statistics you send from your own model (for example an activation norm) and tells you to halt generation when they surge. live tools · copy, paste, send
• CAIN MemVault · CAIN MemVault screens everything an AI agent writes to long-term memory. Known prompt-injection text is quarantined, so it is never recalled later, and every memory record is signed. live tools · copy, paste, send
• CAIN OpenShell · CAIN OpenShell checks each file, network or process action an AI agent wants to take against that agent's sandbox policy, and refuses anything outside it, such as reading /etc/shadow or an SSH key. live tools · copy, paste, send
• CAIN PathCorrection · PathCorrection watches an agent's recent steps for drift: repeated tool loops and wandering away from its goal. When drift passes the tolerance it injects an anchor reminding the agent of its goal and constraints. live tools · copy, paste, send
• CAIN PolicyForge · PolicyForge writes an OPA Rego policy from what an agent actually did, and checks whether that policy would give the agent more permission than its approved baseline. A widening is flagged. live tools · copy, paste, send
• CAIN Proof-Carrying Code · CAIN Proof-Carrying Code checks code an AI agent wants to run before it runs: it must parse, import only allowed modules and contain no dynamic execution (eval, exec, os.system). Pass and it is certified; fail and it is refused. live tools · copy, paste, send
• CAIN RIG · CAIN RIG gives every AI agent its own cryptographic identity and hands it single-use, short-lived capability tokens for one action on one resource. A token works once, only for the agent holding the matching private key, and only for the action it was issued for. live tools · copy, paste, send
• CAIN SCITT Evidence Clearinghouse · The SCITT Evidence Clearinghouse records signed statements (for example 'build 4812 passed its tests') in an append-only log and returns a receipt with a Merkle inclusion proof that anyone can re-check. live tools · copy, paste, send
• CAIN Speculative Fast-Forward · Speculative Fast-Forward evaluates an agent's candidate plans before any of them runs and commits only branches whose every step is safe; a plan containing a destructive step is never committed. live tools · copy, paste, send
• CAIN Steganographic Firewall · The Steganographic Firewall checks what an AI agent sends out for hidden channels: it strips zero-width characters and blocks base64-encoded credentials before they leave. live tools · copy, paste, send
• CAIN Swarm Cartel Detector · The Swarm Cartel Detector reads who delegates to whom in a group of AI agents and flags circular delegation rings, where agents pass authority around in a loop to hide where it came from. live tools · copy, paste, send
• CAIN Swarm Guard · CAIN Swarm Guard is a one-line Python decorator around an AI agent's tools. Before a tool runs it checks an allowlist, the arguments and the risk; a risky call stops before the tool body is reached. When one agent hands work to another, the second can only receive equal or less authority. live tools · copy, paste, send
• CAIN SwarmGossip · SwarmGossip checks claims that AI agents pass to each other. A claim whose originator signature does not verify is isolated, so one agent's made-up 'fact' doesn't spread through the swarm. live tools · copy, paste, send
• CAIN VIGIL · CAIN VIGIL checks each step an AI agent proposes against its stated goal and blocks steps that look destructive or like data exfiltration, or that change things without a justification. A destructive step is pruned, never run. live tools · copy, paste, send
• CAIN zkAgent · zkAgent proves that every step of an agent's run stayed under a public risk ceiling without revealing the actual risk scores, using Pedersen-commitment range proofs. A step over the ceiling cannot be proven. live tools · copy, paste, send
• CAINBudget · CAINBudget gives every AI agent a hard spending limit. Each spend is checked against it first: within the limit is ALLOW, one cent over is DENY, with a signed receipt either way. live tools · copy, paste, send
• CainDrift · CainDrift compares an AI system's answers before and after a model change and flags the answers whose meaning changed (yes became no, numbers changed, 'not' disappeared). live tools · copy, paste, send
• CAINPay · CAINPay sets a spending policy for an AI agent (limits, blocked merchants) and approves or declines each payment against it, with a signed receipt and a tamper-evident ledger. live tools · copy, paste, send
• Capability Lease · Capability Lease lets one AI agent hand part of its permissions to another, and guarantees the hand-off can only shrink: never a new permission, never a bigger budget. live tools · copy, paste, send
• Contamination Scanner · Contamination Scanner tells you whether an AI model memorized the test you are scoring it on, which would make its score meaningless. live tools · copy, paste, send
• Content Provenance · Content Provenance attaches a signed 'made by AI' label to a piece of content. Change one character and the label no longer matches. live tools · copy, paste, send
• Crypto Audit · Crypto Audit reads Python code and finds broken or misused encryption: outdated algorithms, switched-off security checks, keys written into the code. live tools · copy, paste, send
• Data-Exfiltration Killswitch · The Data-Exfiltration Killswitch is an emergency stop: when an AI agent suddenly sends out an abnormal amount of data, it cuts the agent off within milliseconds and writes a signed record. live tools · copy, paste, send
• EU AI Act / ISO 42001 Evidence Notary · The Evidence Notary keeps track of which evidence you have supplied for each ISO 42001 and EU AI Act clause, hashes it and signs the report. A clause with no evidence says NOT_ASSESSED. It never says you comply. live tools · copy, paste, send
• EU AI Act Classifier · Describe an AI system in one sentence; the EU AI Act Classifier tells you its risk tier under the EU AI Act, which phrase triggered it, and the article involved. live tools · copy, paste, send
• FrontierGuard · FrontierGuard is a quick screen of whether an AI model developer is likely covered by rules written for the biggest ('frontier') AI developers, from compute spend, revenue and training size. live tools · copy, paste, send
• LeakGuard · LeakGuard scans text for passwords, API keys, card numbers and personal data before it is sent to an outside AI tool. live tools · copy, paste, send
• Lethal Trifecta Guard · Lethal Trifecta Guard stops an AI agent session the moment it combines three things that together leak data: untrusted content, private data, and a way to send data out. live tools · copy, paste, send
• MCP Attestor · MCP Attestor checks an AI tool two ways: it reads the tool's description, then actually runs the tool in a sealed sandbox to see whether it does what it claims. live tools · copy, paste, send
• MCP Security Scanner · MCP Security Scanner reads the tools an AI agent connects to and quarantines the poisoned ones (hidden instructions, credential grabs, invisible characters) before the agent ever sees them. live tools · copy, paste, send
• MCPGate Boundary Appliance · MCPGate Boundary Appliance is a gate you run yourself between AI agents and their MCP tool servers. Every tool call is evaluated before it is forwarded; a call that is not explicitly allowed does not reach the tool. live tools · copy, paste, send
• MCPGate Guard Stack · The MCPGate Guard Stack puts several guards in front of every tool an AI agent calls (skill scope, memory firewall, delegation checks, supply-chain checks and a kill-switch for bulk exports). Ordinary work goes through; each attack is stopped by a different guard, and refused calls never reach the tool. live tools · copy, paste, send
• MemGuard · MemGuard checks what an AI agent is about to store in long-term memory. A known attack is blocked. Anything that might instruct the agent later is held for a person to review, rather than trusted. live tools · copy, paste, send
• Memory Authority Firewall · Memory Authority Firewall makes sure an AI agent's memory can inform it but never give it permission to do anything. live tools · copy, paste, send
• MemoryMesh · MemoryMesh is long-term memory for AI agents with real keyword search. Text written like a known prompt-injection command is quarantined when stored, so search never hands it back to the agent. live tools · copy, paste, send
• NexusMind · NexusMind is one shared fact store for a company's AI agents. Every fact keeps its source, and when two agents record facts that contradict each other, the conflict is flagged instead of the old fact being silently overwritten. live tools · copy, paste, send
• OmegaRouter · OmegaRouter picks which AI model handles each request by your own rules (capabilities, quality bar, cost, speed), explains the choice, lists fallbacks and stops sending traffic to a model that keeps failing. live tools · copy, paste, send
• Payment Mandate Guard · Payment Mandate Guard checks every purchase an AI shopping agent tries to make against what a person actually approved: the store, the items, who gets paid, how much, and until when. live tools · copy, paste, send
• Physical AI & L5 Autonomy Guards · The Physical AI & L5 Autonomy Guards are software checks for AI that moves things: drones, self-driving vehicles, factory robots, vehicle fleets and defense systems. Each guard stops a command that breaks its rules (no-fly zones, operating limits, spoofed sensors, unsafe speed near people, missing human decision, forged or replayed commands). live tools · copy, paste, send
• PQC Hybrid Signing · PQC Hybrid Signing seals every record twice: once with today's standard method and once with a method designed to resist future quantum computers. live tools · copy, paste, send
• Self-Modification Firewall · Self-Modification Firewall stops an AI agent from changing the things that control it: its shutdown switch, its monitoring, its scoring, its own permissions. live tools · copy, paste, send
• Shadow AI Scanner · Shadow AI Scanner reads a codebase and lists every place it uses AI that nobody declared: AI library imports and direct calls to AI services. live tools · copy, paste, send
• Skill Memory Sandbox · The Skill Memory Sandbox encrypts each AI skill's memory under its own key, so one skill (say, a smart-home add-on) cannot read another's secrets (say, a password vault), even when it runs in the same agent. live tools · copy, paste, send
• Supply-Chain Guard · Supply-Chain Guard stops AI coding agents from installing fake or malicious software packages and from changing the settings that keep them in check. live tools · copy, paste, send
• Swarm Collusion Defense · Swarm Collusion Defense checks groups of AI agents for loops, fake agreement and fake chains of authority, because agents agreeing with each other is not permission. live tools · copy, paste, send
• Swarm Fleet Quarantine · Swarm Fleet Quarantine stops one rogue AI agent, a whole swarm, or every agent at once, and every worker process sees the stop through one shared store. Each stop is signed; lifting it needs two different named people, and the quarantined agent can never approve its own release. live tools · copy, paste, send
• TEE Verifier · TEE Verifier checks that an AI workload really runs inside a hardware-sealed computer (AWS Nitro or AMD SEV-SNP) by checking the chip maker's signed report against the maker's published root keys. live tools · copy, paste, send
• Trust State Engine · Trust State Engine keeps a trust level for each AI agent built only from evidence. After a policy violation, the agent's high-risk actions are denied, with an explanation naming the event. live tools · copy, paste, send
• VeritasEngine · VeritasEngine checks an AI agent's output with plain, repeatable rules (schema, arithmetic, file fingerprint, quotes against the source, forbidden text). One failed check and the whole output is REFUTED. live tools · copy, paste, send
• WORM Evidence Notary · WORM Evidence Notary turns a list of AI decision records into a signed evidence package: each record is hash-chained, bound into a Merkle root and signed with Ed25519. Change, delete or reorder a record and verification fails. live tools · copy, paste, send

FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.