Promote Sign in Sign up

AI security 2026 Rank: #2218

OWASP Agent Memory Guard ✓

Guard your AI agent's memory from security threats.

216 views 5 saves Unknown
Memory Protection AI Security Memory Poisoning
More categories (4)
OWASP Context Poisoning AI Agents AI security
Visit website Sign in to save

Verdict

A quick read before you dig into the full page.

Best if
Guard your AI agent's memory from security threats
Skip if
Complex setup and configuration
Pick this when
Open-source software

Considering AI Security Gateway instead? Compare side by side

Overview

OWASP www-project-agent-memory-guard is an open-source software tool specifically designed to provide robust security for AI agents against memory-based attacks.

It aims to prevent AI agents from getting weaponized through their inherent memory. This tool offers runtime defense functionalities that can detect memory poisoning even after a system context reset.

This resolves a serious security problem as modern AI agents persist memory across sessions, and anything committed to that memory becomes a privileged input on the agent's next operational turn.

This kind of memory poisoning can lead to compromised instructions, data exfiltration, or hijacked tool calls. These attacks can persist even after context resets because the memory itself remains uncompromised.

www-project-agent-memory-guard operates by sitting between the agent and its memory store, screening every operation through a detector pipeline and a declarative policy.

This OWASP Foundation project specifically addresses ASI06 (Memory & Context Poisoning). It was created and led by Vaishnavi Gudur, with co-leader Anshul Rajkumar.

OWASP Agent Memory Guard website

Access & connections

Where you can use OWASP Agent Memory Guard.

Platforms

Web

Capabilities

What this tool accepts, produces, and is strongest at.

Input

  • Code

Output

  • Code

Pros & cons

Pros

  • Open-source software
  • Prevents memory-based attacks
  • Offers runtime defense
  • Detects memory poisoning post-reset
  • Secures privileged input
  • Mitigates compromised instructions
  • Prevents data exfiltration
  • Avoids hijacked tool calls
Show 30 more
  • Operational with system context reset
  • Screens operational pipeline
  • Supports declarative policy
  • Addresses Memory & Context Poisoning (ASI06)
  • Created by security leaders
  • Memory Guard functionality
  • Memory store screening
  • Registry Integrations
  • Improvement on traditional defenses
  • Ships with GuardedChatMessageHistory
  • Features Policy enforcement
  • Structured SecurityEvent emissions
  • Snapshot point-in-time security
  • Supports LangChain Integrations
  • Framework-agnostic MemoryStore compatibility
  • Threat detection feature
  • Detects protected-key modifications
  • Advanced tampering detection
  • YAML-defined rules mapping
  • Scalable language-chain middleware
  • Block, quarantine, allow, redact actions
  • Low latency (59 s median)
  • 100% precision on threat detection
  • Detection across multiple threat categories
  • Runs locally on device
  • Minimal setup required
  • No API keys needed
  • No external calls required
  • Supports memory lifecycle governance
  • Source-class provenance for writes

Cons

  • Limited integrations
  • Potential for false positives
  • Requires technical understanding
  • No stated compatibility
  • No multi-lingual support
  • Undisclosed performance impact
  • Complex setup and configuration
  • No GUI
Show 2 more
  • policy definition needed
  • updates could cause glitches

Compare alternatives

Similar tools in the catalog — with a clear reason to open each one.

Why look Offers a public API

Full comparison

Free + from $10/yr

Full comparison

Why look Offers a public API

Full comparison

Why look Higher community rating

Full comparison

No pricing

Full comparison

No pricing

Full comparison

Releases 1

Initial release August 17, 2026 Current
Initial release of OWASP Agent Memory Guard.

FAQ 12

What is the main purpose of OWASP www-project-agent-memory-guard?
The main purpose of OWASP www-project-agent-memory-guard is to provide robust security for AI agents against memory-based attacks. It is designed to prevent AI agents from being weaponized through their inherent memory.
How does www-project-agent-memory-guard protect against memory-based attacks on AI agents?
www-project-agent-memory-guard operates to protect against memory-based attacks on AI by acting as an intermediary between the agent and its memory storage. Every operation is screened through a detector pipeline and declarative policy, enabling it to detect any potential security threats.
How can I integrate www-project-agent-memory-guard into my existing system?
Integration into an existing system can be achieved through the use of a set of built-in functions provided by www-project-agent-memory-guard. The tool provides a range of developer tools and workflows, allowing it to be integrated into most software development pipelines, regardless of the specific platform or language used.
What types of attacks does www-project-agent-memory-guard protect against?
www-project-agent-memory-guard protects against memory and context poisoning, compromised instructions, data exfiltration, and hijacked tool calls. These are types of attacks that can persist even after context resets and can potentially lead to major security issues.
Does www-project-agent-memory-guard offer protection both during and outside of runtime?
Yes, www-project-agent-memory-guard offers protection both during and outside runtime. It comes with runtime defense functionalities that continuously work to detect any instances of memory poisoning and other potential attacks.
Can www-project-agent-memory-guard detect memory poisoning after a system context reset?
Yes, www-project-agent-memory-guard can detect memory poisoning even after a system context reset, providing a robust defence mechanism against persistent attacks that leverage the inherent memory of an AI agent.
What types of security threats can www-project-agent-memory-guard protect my AI agent's memory from?
www-project-agent-memory-guard can protect an AI agent's memory from various security threats such as memory poisoning, compromised instructions, data exfiltration, hijacked tool calls, and any attack that can leverage the agent's memory as a tool for weaponization.
Who created www-project-agent-memory-guard?
www-project-agent-memory-guard was created and led by Vaishnavi Gudur, with co-leader Anshul Rajkumar.
What is the relationship between OWASP and www-project-agent-memory-guard project?
www-project-agent-memory-guard is a project by the OWASP Foundation. The foundation's goals align with the project's - to make software security visible, so individuals and organizations are able to make informed decisions about true software security risks.
Can www-project-agent-memory-guard protect against compromised instructions, data exfiltration, and hijacked tool calls?
Yes, www-project-agent-memory-guard can protect against compromised instructions, data exfiltration, and hijacked tool calls. It achieves this by screening every operation through a detector pipeline and a declarative policy to identify and prevent such threats.
What does 'runtime defense functionalities' mean in reference to www-project-agent-memory-guard’s features?
'Runtime defense functionalities' refer to www-project-agent-memory-guard’s ability to actively scan, detect and prevent threats during the operation of the AI agent. This ensures that the AI agent is protected from dangerous inputs and breaches in real time.
How does www-project-agent-memory-guard deliver policy screening?
www-project-agent-memory-guard delivers policy screening by using a declarative policy that checks every operation that interacts with the agent's memory. This means that all data is screened against this policy to ensure it doesn't present any security threats to the system.
OWASP Agent Memory Guard
Visit website