{"data":{"slug":"agent-wallet-threat-model","title":"Agent Wallet Threat Model: Keys, Sessions, Budgets, and Receipts","tags":["AI Privacy","Wallets","OPSEC"],"desc":"A practical threat model for AI-agent wallets: key custody, prompt injection, sessions, budgets, receipts, approval policies, wallet reuse, and privacy logs.","time":"~5 min read","date":"2026-07-05","lastReviewed":"2026-07-13","contentKind":"brief","editorialStatus":"ok","providerSlugs":[],"primaryProviderSlugs":[],"topic":"Agent signing authority and receipt exposure","relatedArticleSlugs":["ai-tools-permissions-security","ap2-x402-mpp-agent-payment-privacy","virtual-cards-vs-stablecoin-wallets-ai-agents"],"url":"https://cunicula.com/en/articles/agent-wallet-threat-model","answer":"A practical threat model for AI-agent wallets: key custody, prompt injection, sessions, budgets, receipts, approval policies, wallet reuse, and privacy logs.","sources":[{"label":"AP2 v0.2 specification","url":"https://ap2-protocol.org/ap2/specification/"},{"label":"Coinbase x402 docs","url":"https://docs.cdp.coinbase.com/x402/welcome"},{"label":"Stripe machine payments docs","url":"https://docs.stripe.com/payments/machine"},{"label":"Ethereum privacy documentation","url":"https://ethereum.org/en/privacy/"},{"label":"OWASP prompt injection guidance","url":"https://genai.owasp.org/llmrisk/llm01-prompt-injection/"}],"section_headings":["Main assets","Main failure modes","Safer architecture","Minimum controls"],"related_articles":[{"slug":"ai-tools-permissions-security","title":"AI Tools: Permissions, Packages and Security Checks","tags":["AI Privacy","Security","Guide"],"lastReviewed":"2026-11-02","contentKind":"guide","editorialStatus":"ok","desc":"How to check what AI connectors and local tools can read, limit credential exposure, recognise prompt injection and verify running security updates.","time":"~4 min read","date":"2026-11-02","providerSlugs":[],"primaryProviderSlugs":[],"topic":"AI tool permissions and package compromise","relatedArticleSlugs":["agent-wallet-threat-model","private-ai-local-llm","ai-agent-money-privacy-controls"],"url":"https://cunicula.com/en/articles/ai-tools-permissions-security","answer":"How to check what AI connectors and local tools can read, limit credential exposure, recognise prompt injection and verify running security updates.","sources":[{"label":"Greshake and colleagues' research, revised 5 May 2023","url":"https://arxiv.org/abs/2302.12173v2"},{"label":"versioned security guidance","url":"https://modelcontextprotocol.io/docs/2026-07-28/tutorials/security/security_best_practices"},{"label":"Model Context Protocol tool specification","url":"https://modelcontextprotocol.io/specification/2025-06-18/server/tools"},{"label":"Node 22.23.4, released on 21 October 2026","url":"https://nodejs.org/en/blog/release/v22.23.4"},{"label":"Ubuntu's Brotli advisory","url":"https://ubuntu.com/security/CVE-2026-111497"},{"label":"4 August 2026 ChainDrop analysis","url":"https://www.microsoft.com/en-us/security/blog/2026/08/04/chaindrop-supply-chain-compromise-anatomy-self-propagating-worm/"}],"section_headings":["What each connection can read","External text can become an instruction","Packages can run before the assistant starts","Check the update that is actually running"]},{"slug":"ap2-x402-mpp-agent-payment-privacy","title":"AI Agent Payment Protocols Compared","tags":["AI Privacy","Payments","Agent Money"],"desc":"A privacy-first comparison of AP2, x402, and MPP for AI-agent payments: authorization, HTTP 402 challenges, settlement rails, logs, wallets, and metadata exposure.","time":"~5 min read","date":"2026-07-05","lastReviewed":"2026-11-08","contentKind":"brief","editorialStatus":"ok","providerSlugs":[],"primaryProviderSlugs":[],"topic":"Agent payment authorization and settlement protocols","relatedArticleSlugs":["agent-wallet-threat-model","pay-per-crawl-http-402-privacy-risks","score-agent-payment-rail"],"url":"https://cunicula.com/en/articles/ap2-x402-mpp-agent-payment-privacy","answer":"A privacy-first comparison of AP2, x402, and MPP for AI-agent payments: authorization, HTTP 402 challenges, settlement rails, logs, wallets, and metadata exposure.","sources":[{"label":"AP2 security and privacy specification","url":"https://ap2-protocol.org/ap2/security_and_privacy_considerations/"},{"label":"Google AP2 announcement","url":"https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol"},{"label":"Cloudflare MPP compatibility note","url":"https://developers.cloudflare.com/agents/tools/payments/mpp/"},{"label":"Coinbase x402 docs","url":"https://docs.cdp.coinbase.com/x402/welcome"},{"label":"Stripe Machine Payments Protocol","url":"https://stripe.com/blog/machine-payments-protocol"}],"section_headings":["What each layer does","The privacy split","Privacy guidance","Before letting an agent pay"]},{"slug":"virtual-cards-vs-stablecoin-wallets-ai-agents","title":"Virtual Cards vs Stablecoin Wallets for AI Agents","tags":["AI Privacy","Payments","Wallets"],"desc":"A privacy-first comparison of virtual cards and stablecoin wallets for AI-agent spending: KYC, issuer logs, onchain history, revocation, budgets, and merchant fit.","time":"~4 min read","date":"2026-07-05","lastReviewed":"2026-08-17","contentKind":"brief","editorialStatus":"ok","providerSlugs":[],"primaryProviderSlugs":[],"topic":"Virtual cards versus stablecoin agent payments","relatedArticleSlugs":["privacy-com-mcp-spend-control-not-anonymity","stablecoin-freezing-powers","agent-wallet-threat-model"],"url":"https://cunicula.com/en/articles/virtual-cards-vs-stablecoin-wallets-ai-agents","answer":"A privacy-first comparison of virtual cards and stablecoin wallets for AI-agent spending: KYC, issuer logs, onchain history, revocation, budgets, and merchant fit.","sources":[{"label":"Privacy AI agent cards page","url":"https://agents.privacy.com/"},{"label":"Coinbase x402 docs","url":"https://docs.cdp.coinbase.com/x402/welcome"},{"label":"Stripe machine payments docs","url":"https://docs.stripe.com/payments/machine"}],"section_headings":["Virtual-card strengths","Stablecoin-wallet strengths","How to choose","Comparison checklist"]}],"relationships":{"entity_id":"https://cunicula.com/en/articles/agent-wallet-threat-model#article","topic":"Agent signing authority and receipt exposure","providers":[],"topics":[],"related_articles":[{"slug":"ai-tools-permissions-security","title":"AI Tools: Permissions, Packages and Security Checks","url":"https://cunicula.com/en/articles/ai-tools-permissions-security","entity_id":"https://cunicula.com/en/articles/ai-tools-permissions-security#article"},{"slug":"ap2-x402-mpp-agent-payment-privacy","title":"AI Agent Payment Protocols Compared","url":"https://cunicula.com/en/articles/ap2-x402-mpp-agent-payment-privacy","entity_id":"https://cunicula.com/en/articles/ap2-x402-mpp-agent-payment-privacy#article"},{"slug":"virtual-cards-vs-stablecoin-wallets-ai-agents","title":"Virtual Cards vs Stablecoin Wallets for AI Agents","url":"https://cunicula.com/en/articles/virtual-cards-vs-stablecoin-wallets-ai-agents","entity_id":"https://cunicula.com/en/articles/virtual-cards-vs-stablecoin-wallets-ai-agents#article"}]},"content":{"slug":"agent-wallet-threat-model","answer":"A practical threat model for AI-agent wallets: key custody, prompt injection, sessions, budgets, receipts, approval policies, wallet reuse, and privacy logs.","publishedAt":"2026-07-05","reviewedAt":"2026-07-13","sections":[{"heading":"Main assets","paragraphs":["Money is only one asset. An agent payment can expose account identity, purchase intent, wallet history, receipts, API keys, merchant sessions, and the prompts that explain why the payment happened.","Separate payment intent from signing authority. AP2 does this with checkout and payment mandates that other roles verify. The same boundary can be simpler: let the agent prepare a request, then let deterministic code or a person approve it."],"bullets":[]},{"heading":"Main failure modes","paragraphs":["The worst design puts seed material, browser automation, untrusted web content, and open-ended spending authority in one runtime. OWASP warns that indirect prompt injection can arrive through websites or files and trigger unauthorized tool use or commands.","A second failure is receipt sprawl. If receipts, prompts, files, and identity documents land in one store, the payment trail becomes a map of the user."],"bullets":["Prompt injection asks the agent to pay the wrong party.","A merchant session leaks identity across unrelated tasks.","Wallet address reuse links agent tasks on a public ledger.","A broad API key lets the agent create spend surfaces without review."]},{"heading":"Safer architecture","paragraphs":["Use a separate wallet, card, or balance for each task compartment. Keep signing authority outside the browsing and prompt runtime. Put a deterministic policy check between the agent request and the payment tool.","Require human approval for high-risk payments. For repeat low-value work, enforce amount, merchant, frequency, instrument, and expiry limits. AP2 autonomous mandates use signed constraints; x402 instead lets a client construct and send payment data during an HTTP request. Neither protocol removes the need to limit the wallet that the client can use."],"bullets":[]},{"heading":"Minimum controls","paragraphs":[],"bullets":["No seed phrases in the agent runtime.","One wallet, card, or balance per task compartment.","Hard maximum amount before the workflow starts.","Human approval above a tiny threshold.","Receipt logs without prompts, private files, or secrets."]}],"faq":[{"question":"Should an AI agent hold wallet keys?","answer":"Usually no. The safer pattern is for the agent to draft a payment request and for a separate wallet, policy engine, or human to approve and sign."},{"question":"What should an agent payment wallet log?","answer":"Log merchant, amount, task, instrument, approval result, and receipt reference. Avoid storing prompts, private files, seed material, bank data, or unrelated identity data in the same log."}],"sources":[{"label":"AP2 v0.2 specification","url":"https://ap2-protocol.org/ap2/specification/"},{"label":"Coinbase x402 docs","url":"https://docs.cdp.coinbase.com/x402/welcome"},{"label":"Stripe machine payments docs","url":"https://docs.stripe.com/payments/machine"},{"label":"Ethereum privacy documentation","url":"https://ethereum.org/en/privacy/"},{"label":"OWASP prompt injection guidance","url":"https://genai.owasp.org/llmrisk/llm01-prompt-injection/"}],"hasDedicatedSourceLedger":true,"figure":{"id":"agent-wallet-approval","figId":"FIG.05","title":"The agent drafts. It never signs.","alt":"Approval path from agent-drafted payment request through policy check, human approval, and an external signer, with four failure modes and where each is caught.","caption":"The safer architecture from the article: intent and signing authority are separated, with limits enforced before payment.","image":"https://cunicula.com/article-figures/agent-wallet-threat-model/agent-wallet-approval--article-desktop.png","width":1200,"height":475,"sources":["https://ap2-protocol.org/ap2/specification/","https://docs.cdp.coinbase.com/x402/welcome","https://genai.owasp.org/llmrisk/llm01-prompt-injection/"],"semanticData":{"caption":"The safer architecture from the article: intent and signing authority are separated, with limits enforced before payment.","headers":["Failure","Effect","Caught by"],"rows":[["PROMPT INJECTION","asks the agent to pay the wrong party","policy check + approval"],["MERCHANT SESSION LEAK","identity crosses unrelated tasks","one wallet per task compartment"],["WALLET ADDRESS REUSE","links agent tasks on a public ledger","one wallet per task compartment"],["BROAD API KEY","creates spend surfaces without review","hard maximum before the workflow starts"]]}}}},"meta":{"total":1,"api_version":"1.3","powered_by":"cunicula.com","documentation_url":"https://cunicula.com/en/api-docs","machine_index_url":"https://cunicula.com/llms.txt","license":"LicenseRef-Cunicula-Attribution-1.0","license_url":"https://cunicula.com/en/cite","attribution":"Cunicula, \"{name}\", cunicula.com, reviewed {reviewed}, retrieved {retrieved}, {url}","cite_guide_url":"https://cunicula.com/en/cite","dataset_url":"https://cunicula.com/api/v1/dataset","schema_version":"1.1.0","related_total":3,"has_content":true}}