← Frameworks
MANAGE 2.4

NIST AI RMF

NIST AI RMFView in NIST

How MANAGE 2.4 shows up in practice: the mapped risk classes in this atlas, the documented incidents that prove it's real, and the scenarios and controls to learn and defend against it.

Real-world cases

37

Documented incidents, disclosed vulnerabilities and research that illustrate MANAGE 2.4 — latest first, each with sources.

Context Contamination: passive prompt injection poisons LLM security-log analysis16 Jul 2026

A red-team study shows adversaries can hide prompt-injection payloads inside network-log fields (usernames, URLs, user-agents) that fire when a SOC analyst asks an LLM to triage the logs — reportedly reaching up to 88.2% success at concealing malicious activity or exfiltrating data, turning the audit trail itself into the injection channel.

Cursor 'DuneSlide' — indirect prompt injection escapes the IDE sandbox to zero-click RCE (CVE-2026-50548 / CVE-2026-50549)01 Jul 2026

Cato AI Labs disclosed two critical (CVSS 9.8) zero-click flaws in Cursor's coding agent where a single instruction hidden in content the agent reads — an MCP tool response or a web-search result — escapes the editor's terminal sandbox and runs OS-level commands with no click or approval.

Amazon Q Developer auto-loads workspace MCP configs, enabling zero-click AWS credential theft (CVE-2026-12957)26 Jun 2026

Wiz Research found Amazon Q's VS Code extension auto-loaded MCP server definitions from a repo's .amazonq/mcp.json with no consent or workspace-trust prompt; opening a booby-trapped repository silently spawned attacker-controlled processes that inherited the developer's full environment and could stream live AWS session credentials out to an attacker.

TOCTOU perceive-then-act race in computer-use agents (Claude Computer-Use)25 Jun 2026

Johann Rehberger showed a time-of-check/time-of-use race in GUI/computer-use agents: the screen can change after the agent captures its screenshot but before its click lands, so a benign-looking 'Continue' button silently resolves to an Outlook 'Send'. Anthropic tracked the issue and Cowork now revalidates pixels before acting.

SearchLeak — Microsoft 365 Copilot one-click data theft (CVE-2026-42824)15 Jun 2026

A single malicious link reportedly turned Copilot Enterprise Search's URL query parameter into an executable prompt, exfiltrating emails, MFA codes and files via a Bing image-search side channel.

Agentjacking — hijacking AI coding agents via Sentry error reports (Tenet Security)12 Jun 2026

Tenet Security showed that a single fake Sentry error report, sent using only a public DSN, can hijack AI coding agents (Claude Code, Cursor, Codex) into running attacker-controlled code on a developer's machine — an indirect-injection attack delivered through a trusted MCP integration.

Poisoning Claude Code: one GitHub issue hijacks the claude-code-action CI supply chain01 Jun 2026

GMO Flatt Security's RyotaK showed that a single attacker-opened GitHub issue could indirect-prompt-inject Anthropic's claude-code-action CI agent — whose permission check reportedly trusted any "[bot]" actor — coaxing Claude to leak CI secrets and OIDC tokens, gain repository write access, and potentially poison the shared action that downstream repos pull via a floating tag.

Meta AI support bot tricked into hijacking Instagram accounts31 May 2026 – 01 Jun 2026

Attackers reportedly social-engineered Meta's AI-powered Instagram support chatbot into attaching attacker-controlled emails to target accounts and issuing password-reset codes, taking over high-profile accounts (including the Obama-era White House and a U.S. Space Force CMSgt) without the owner's email or any MFA prompt.

ChatGPhish — ChatGPT web-summary rendering turned into a phishing surface29 May 2026

Attacker-controlled Markdown hidden in a public web page is reportedly rendered by ChatGPT's summarization feature as trusted assistant output — spoofed OpenAI alerts, phishing links, QR codes, and tracking pixels.

ClaudeBleed — co-resident Chrome extensions coerce Claude for Chrome into reading Gmail/Docs/Calendar21 May 2026 / 14 Jul 2026

Manifold Security reported that any co-resident browser extension could weaponize Claude for Chrome — dispatching synthetic clicks the agent accepted without checking Event.isTrusted, and loading its side panel with ?skipPermissions=true — to make the AI read the victim's Gmail, Docs and Calendar; reportedly still unpatched across eight releases (CVSS up to 9.6, per the researchers).

Grok + Bankrbot Morse-code prompt injection drains on-chain wallet04 May 2026

An X user escalated Grok's on-chain wallet via a Bankr Club NFT, then sent a Morse-code instruction Grok auto-decoded and relayed to the autonomous agent Bankrbot — moving ~3B tokens (reportedly ~$150K-$200K) with no secondary verification.

PyTorch Lightning PyPI compromise (Mini Shai-Hulud / TeamPCP)30 Apr 2026

Malicious 'lightning' PyPI releases (reportedly 2.6.2 and 2.6.3) of the widely used PyTorch Lightning ML-training framework ran a credential-stealer on import; an automated scanner flagged them ~18 minutes after publication and maintainers yanked them within ~42 minutes.

+ 25 more via the mapped risk pages above.

Browse all real-world cases →

Practise it — interactive scenarios

📧The Email That Gave Orders

A support email hides instructions — and the assistant obeys them

🕵️Lies in the Loop

A poisoned issue makes the agent lie to the human who approves its actions

🏭Poisoning the Agent Factory

Compromise the pipeline that builds agents, and every new worker is born malicious

🪤The Bug Report That Ran Code

A fake Sentry error report hijacks a developer's coding agent into running a shell command

📼The Compromised Flight Recorder

The forensic record is itself the attack surface — an agent's log is poisoned, then quietly rewritten

👁️The Invisible Webpage Command

A shopping page tells the agent to do something the user never asked for

🕵️The Logs That Lied

An attacker plants prompt injection in the audit trail — so the LLM that hunts them erases the evidence

🧠The Memory That Wouldn't Die

A single poisoned document plants a standing instruction that survives every reset

📡The Message in Morse

Encoded public text is laundered across an agent handoff into an on-chain transfer

🖼️The Picture That Whispered

A screenshot that's harmless at full size becomes an order once the system shrinks it

🥸The Uninvited Agent

A forged peer registers on the agent directory — and the planner enlists it

🛡️The Watcher Watched

The eval gate that was supposed to catch the agent is itself the thing being attacked

🪪The Worker Who Spoke for the Boss

A poisoned web page hijacks a research agent — and the planner acts on its behalf

🖱️What You Click Is Not What You Get

A GUI agent clicks 'Continue' — but the screen moved, and it lands on 'Send'

🖼️Zero-Click Leak by Picture

An inbox summary quietly ships a secret to an attacker's server

Controls & guardrails that address this

273 proposed

Guardrails across the risks mapped to MANAGE 2.4, grouped by control function. Filter by control category below.

Control category
Preventive · 15
Role-based access controls

Design the system prompt architecture with privilege separation and trust tier definitions at design stage.

Jailbreak detection

Implement input sanitisation and injection detection filters covering known injection patterns and privilege escalation attempts.

Lifecycle stages3 – Onboarding, Build & Review4 – Deployment
Spotlighting of untrusted content via delimiting, datamarking and encoding

Wrap all untrusted content in random delimiters and datamarking; instruct the model never to execute instructions inside the marked region. Gate release on injection eval results.

source: Microsoft 'Spotlighting' technique (Hines et al. 2024); OWASP Top 10 for LLM Apps LLM01:2025 Prompt Injection (segregate external content)
Lifecycle stage3 – Onboarding, Build & Review
Dedicated injection-detection classifier on all inbound untrusted content and outbound actions

Benchmark the classifier on a labelled injection corpus and tune the decision threshold. Sign off the operating point before deployment.

source: MITRE ATLAS AML.M0015 (Adversarial Input Detection); OWASP Top 10 for LLM Apps LLM01:2025 Prompt Injection; NIST AI RMF MEASURE 2.7
Lifecycle stages3 – Onboarding, Build & Review4 – Deployment5 – Usage, Monitoring & Change
Multimodal input-fidelity check: show/verify the model-delivered (post-downscale) image and avoid silent lossy resampling✚ proposed

Before inference, render a preview of the exact image (and dimensions) the model will receive after preprocessing, and either avoid silent downscaling or constrain ingest dimensions — so an attacker cannot hide a payload that only becomes legible after resampling. Closes the inspected-vs-delivered gap that text-based injection filters miss.

source: Case study: anamorpher-image-scaling-injection (Trail of Bits — Morozova & Hussain, 21 Aug 2025)
Lifecycle stage3 – Development & Build
Instruction-hierarchy-trained model selection with role-precedence injection evals✚ proposed

Select or fine-tune the foundation model for a trained instruction-hierarchy prior so system-prompt directives intrinsically outrank user- and tool-originated instructions, and gate release on role-precedence override evals quantifying the residual (behavioural, non-enforced) flip rate.

source: Interactive-control reconciliation: ctrl-instruction-hierarchy (partial coverage)
Lifecycle stage3 – Onboarding, Build & Review
Instruction hierarchy / privileged system promptinteractive

Training the model to treat the app's standing instructions as more authoritative than anything a user or document says.

Delimiting / spotlighting of untrusted contentinteractive

Clearly fencing off outside text — 'everything between these marks is just data, not instructions' — so the model is less likely to obey it.

Ingestion sanitisation & source allowlistinginteractive

Cleaning documents as they enter the library — stripping hidden text and active instructions — and only ingesting from trusted places.

Egress allowlisting & DLP on tool argumentsinteractive

Controlling where the AI can send data, so secrets can't be quietly shipped to a stranger's address or website.

Human-in-the-loop approval on high-risk actionsinteractive

Pausing to ask a person before doing anything big or hard to undo — sending money, deleting data, emailing customers.

Memory write validation, provenance & reviewinteractive

Being careful about what gets saved to long-term memory, labelling where it came from, and letting users see and delete their memories.

Per-agent identity & taint-marked messagesinteractive

Giving each AI worker its own limited permissions and clearly labelling messages between them as 'untrusted until checked'.

Inter-agent authentication & admission controlinteractive

Give every AI agent a verifiable ID badge, keep a guest list of which agents are allowed on the team, and check the badge on every message — so an impostor or an uninvited agent can't be trusted.

Detective · 10
Vulnerability assessment

Conduct a prompt injection threat assessment at design stage covering all input vectors (user, tool, external data).

Penetration testing

Penetration test all prompt injection pathways in the system. Prioritise external tool and document ingestion channels.

Continuous adversarial prompt-injection red teaming with regression suite in CI/CD

Build the versioned injection corpus into CI/CD as a pre-release gate. Baseline attack success and sign off the release threshold.

source: NIST AI RMF MANAGE 2.2 / MEASURE 2.7; MITRE ATLAS AML.M0019 (Red Teaming); OWASP Top 10 for LLM Apps LLM01:2025 (adversarial testing)
Lifecycle stages3 – Onboarding, Build & Review5 – Usage, Monitoring & Change
Materialised model-context audit capture (post-truncation prompt, retrieved and tool content) with read-time redaction✚ proposed

Log the exact post-truncation context the model ingested, including retrieved and tool-returned content rather than only user input, with redaction applied at read time, so indirect injection via that content is forensically visible.

source: Interactive-control reconciliation: ctrl-logging (partial coverage)
Lifecycle stage5 – Usage, Monitoring & Change
Input guardrail / injection classifierinteractive

A screen that reads incoming messages and blocks obvious attacks or banned topics before the model sees them.

Provenance & content signinginteractive

Keeping a label on every document saying where it came from, so you can tell trusted company docs from random web text.

Full-trace audit logginginteractive

Recording everything — questions, documents fetched, actions taken — so you can investigate when something goes wrong.

Memory anomaly detection & quarantineinteractive

Watching for strange new memories — like instructions that suddenly appear — and holding them aside until checked.

Loop/cost circuit-breakers & consistency checksinteractive

Automatic stop-switches when AIs get stuck in loops, burn too much money, or start disagreeing with each other.

Corrective · 3
Red teaming

Conduct comprehensive prompt injection red team exercises (direct, indirect, multi-turn) before deployment.

Data/instruction trust-boundary enforcement with capability gating on injection-reachable tools

Classify content sources into trust tiers at design; place privileged tools behind a tier requiring user-originated intent or human approval. Sign off the trust-tier map before build.

source: Google DeepMind CaMeL (2025); OWASP Agentic AI Threats & Mitigations (tool misuse / compromise); NIST SP 800-53 AC-6 Least Privilege
Lifecycle stages1 – Use Case Context & Design3 – Onboarding, Build & Review
Spotlighting of untrusted content via delimiting, datamarking and encoding

Re-run injection evals on every template change and periodically against new attack techniques. Manage the spotlighting wrapper under change control.

source: Microsoft 'Spotlighting' technique (Hines et al. 2024); OWASP Top 10 for LLM Apps LLM01:2025 Prompt Injection (segregate external content)
Lifecycle stage5 – Usage, Monitoring & Change
Open the Control Library →

AI RiskAtlas is an educational model of how GenAI & agentic systems work and fail. Architectures and payloads are illustrative and simplified for learning — not operational guidance. Real-world cases are summarised from public reporting.

Sources & further reading →·Built by Shi Yuan ↗