AI agents call memory_search to retrieve information from Openai without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
This tool retrieves or queries data (ChatGPT memory entries) in response to keywords without side effects. The redaction of PII indicates defensive design for a read operation. It falls squarely into the 'Read' category. Severity is low because search results on user memories pose minimal risk compared to write, execute, or destructive actions, though sensitive information exposure is mitigated by redaction.
From the tool's definition Tool name 'memory_search' and description 'Keyword search over ChatGPT memories. Returns matching entries' indicate a read-only query operation. No modification, deletion, or execution of external operations is described.
Attacks that exploit this kind of access
Keyword search over ChatGPT memories. Returns matching entries (PII redacted). It is categorised as a Read tool in the Openai MCP Server, which means it retrieves data without modifying state.
Register the Openai MCP server in PolicyLayer and add a rule for memory_search: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Openai. Nothing to install.
memory_search is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the memory_search rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for memory_search. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
memory_search is provided by the Openai MCP server (robotlearning123/gpt2agent). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Every MCP server has a record like this.
Type a name, get the same breakdown: verified identity, auth posture, risk grade, capabilities, recommended policy.
Teams ship this data inside their own products. See what a licence covers →