An open-source project called MGnarf launched as an execution security gate that blocks AI agents from performing protected actions without independently verified authorization, so agents cannot approve their own work. A companion project, MGnarfTest, exists to adversarially attack the gate. The maintainer is recruiting developers and security researchers to try to break it.
WarrantKit is an open-source Python layer that restricts what AI agents can do at runtime using a warrant/lease permission model with fail-closed enforcement. It targets developers building agents who need verifiable sandboxing rather than prompt-based safety. The author is soliciting adversarial testing to find weaknesses beyond their own test coverage.
WarrantKit is an open-source Python layer that restricts what AI agents can do at runtime using a warrant/lease permission model with fail-closed enforcement. It targets developers building agents who need verifiable sandboxing rather than prompt-based safety. The author is soliciting adversarial testing to find weaknesses beyond their own test coverage.
The Agaemon team is recruiting developers building agents in finance, trading, treasury, DeFi, or blockchain to integrate and test its governance layer, which sits between an agent's proposed actions and their execution. The layer evaluates each proposed action against an authorization/policy boundary and returns a decision with evidence. The team is offering sandbox integrations and says it wants critical feedback.
The Agaemon team is recruiting developers building agents in finance, trading, treasury, DeFi, or blockchain to integrate and test its governance layer, which sits between an agent's proposed actions and their execution. The layer evaluates each proposed action against an authorization/policy boundary and returns a decision with evidence. The team is offering sandbox integrations and says it wants critical feedback.
Synentra is an open-source gateway that sits between AI agents and APIs, evaluating requests through intent, risk/trust, policy, and human approval rather than only identity and endpoint access. It addresses the gap where an authenticated agent is allowed to call an endpoint but the specific action (e.g., DELETE /customers/123) should not proceed. The developer is seeking other developers to test the approach.
Synentra is an open-source gateway that sits between AI agents and APIs, evaluating requests through intent, risk/trust, policy, and human approval rather than only identity and endpoint access. It addresses the gap where an authenticated agent is allowed to call an endpoint but the specific action (e.g., DELETE /customers/123) should not proceed. The developer is seeking other developers to test the approach.
Agaemon is a governance and decision layer that sits between AI agents and their execution paths, evaluating whether a proposed action is authorized before it proceeds. Agents propose actions, Agaemon authorizes them, execution runs, and results are verified with evidence produced. The team is seeking developers in finance, trading, treasury, DeFi, and blockchain to integrate and test it with real agent workflows.
Agaemon is a governance and decision layer that sits between AI agents and their execution paths, evaluating whether a proposed action is authorized before it proceeds. Agents propose actions, Agaemon authorizes them, execution runs, and results are verified with evidence produced. The team is seeking developers in finance, trading, treasury, DeFi, and blockchain to integrate and test it with real agent workflows.
AgentGuard is a small permission layer that sits between an AI agent and its tools, checking each call before execution. It verifies tool allowance in the current agent state, argument policy compliance, and unknown/unsafe states, and ensures denied functions remain untouched. It targets developers building LLM agents who need guardrails on tool use.
AgentGuard is a small permission layer that sits between an AI agent and its tools, checking each call before execution. It verifies tool allowance in the current agent state, argument policy compliance, and unknown/unsafe states, and ensures denied functions remain untouched. It targets developers building LLM agents who need guardrails on tool use.