Forged at USCYBERCOM. We replace "Detect & Respond" with Engineered Certainty.
Architects of:
🛡️ Warden
🔗 Digital Shield
🤖 AI SAFE²
👇 Get the Blueprint
The Death of Governance & Compliance in an AI Era..
This statement is obvious for some, obscure for others and for many it is down right confusing.
Most "AI Security" advice is focused on governance and compliance guidance...aka how to create an AI policy or program.
It’s vague.
It’s theoretical.
It’s useless to an engineer.
But most of all it misses the point with AI.
We've been lead to believe if...
We document, monitor or detect risk, we can manage it and everything will be OK.
Reality is...
AI risk is executed in milliseconds & Documentation Does NOT STOP Execution.
♟️Strategic Truth
Most AI security advice from a policy standpoint:
🚫Optimizes for regulatory defensibility
🚫Assumes post-event containment
🚫Treats AI as a governance problem
But AI exploitation is an engineering problem.
Thus, until policy mandates preventive, runtime-enforced controls, AI security guidance will remain structurally misaligned with the threat landscape.
This is where we stepped in when we built the architecture standard back in Jun 25.
Been working hard on AI SAFE² ever since but accelerated after the explosion of risk last quarter by turning this holiday season into go-mode.
We Finalized a v2.0 then quickly finalized a v2.1 both are massive upgrades & industry must haves!
Introducing the AI SAFE² Framework (v2.1):
☑️The Universal GRC Standard for Agentic AI & ISO 42001 Compliance.
📂The open-source protocol for governing Agentic AI.
- Dropped our Github Repo
Let's unpack what all this actually means for AI Governance & Compliance moving forward...🧵
Star the repo to show your support👇
The biggest security problem in agentic AI isn't prompt injection.
It's trust.
Today's A2A protocols solve communication.
They don't solve verification.
That creates 5 structural risks:
🔹Agent impersonation
🔹Unbounded delegation chains
🔹Memory poisoning
🔹Tool-level privilege escalation
🔹Non-verifiable audit trails
Think about it:
Agent A → Agent B → Agent C → Tool Execution
Can you prove:
• who initiated the action?
• what authority was delegated?
• whether permissions expanded?
• what memory influenced the decision?
• who is accountable?
For most deployments, the answer is "not cryptographically."
We're building the HTTP era of agent systems.
The next challenge is building the TLS era.
The organizations that solve agent identity, provenance, delegation, and accountability first will have a significant advantage over those that don't.
The biggest security problem in agentic AI isn't prompt injection.
It's trust.
Today's A2A protocols solve communication.
They don't solve verification.
That creates 5 structural risks:
🔹Agent impersonation
🔹Unbounded delegation chains
🔹Memory poisoning
🔹Tool-level privilege escalation
🔹Non-verifiable audit trails
Think about it:
Agent A → Agent B → Agent C → Tool Execution
Can you prove:
• who initiated the action?
• what authority was delegated?
• whether permissions expanded?
• what memory influenced the decision?
• who is accountable?
For most deployments, the answer is "not cryptographically."
We're building the HTTP era of agent systems.
The next challenge is building the TLS era.
The organizations that solve agent identity, provenance, delegation, and accountability first will have a significant advantage over those that don't.
We just shipped the sovereign layer agentic AI has been missing.
NEXUS-A2A v0.3
Open source. Apache 2.0.
189 tests. Zero external dependencies in the core suite.
The fundamental problem with today's agent ecosystem isn't MCP, ACS, A2A, LangChain, CrewAI, or orchestration frameworks.
The problem is that most systems can authenticate an application, but cannot reliably answer:
• Which agent actually initiated this action?
• What authority was delegated to it?
• How many delegation hops occurred?
• Was its memory manipulated since the last session?
• Who owns responsibility for its actions?
That is a structural governance gap, not a configuration problem.
NEXUS closes that gap with a cryptographic sovereign layer that wraps existing agent infrastructure without requiring changes to MCP servers, ACS Guardians, LangChain, CrewAI, n8n, or other orchestration platforms.
Core controls include:
→ DID + SPIFFE identity on every agent-to-agent message
→ Verifiable Constraint Chains (VCC) that narrow scope at every delegation hop
→ Memory Vaccine drift detection to block memory injection before persistence
→ Guardian enforcement of arguments outside the agent process
→ NOR receipts that create signed, OCSF-mapped audit records for every action
The result is governance, provenance, and accountability that travel with the request itself.
Using the AI SAFE² v3.0 framework:
• Typical MCP deployments score ~11/25
• ACS implementations score ~14/25
• Current A2A implementations score ~8/25
• NEXUS-A2A v0.3 scores 24/25
The remaining point requires production-scale behavioral analytics over extended operational horizons.
We believe every fleet operating 50+ autonomous agents should satisfy six invariants:
I-1: Cryptographic identity at every boundary
I-2: Scope narrows at every delegation hop
I-3: Memory provenance on every cross-session write
I-4: Physical kill switch registered with sub-second propagation
I-5: Named human owner-of-record for every agent
I-6: Behavioral drift treated as a security event
You can verify your own environment today:
pip install nexus-a2a-sdk
Run the compliance checker and receive a control-by-control assessment showing exactly where governance gaps exist.
NEXUS-TGC (multi-sovereign governance committee) is now accepting steering nominations through September 1, 2026.
An IETF Internet-Draft for the identity and transport layers is in preparation.
This is not a vendor lock-in strategy.
This is an open protocol designed to make agent identity, delegation, memory provenance, and accountability verifiable across the emerging agentic ecosystem.
We are excited to join Nvidia's Nemotron Coalition of leading AI labs working together to advance open frontier foundation models.
To celebrate we have partnered with @nvidia and @nebiustf to provide 2 free weeks of the new Nemotron 3 Ultra model on the Nous Portal!
The era of "AI Tool Calling" is ending.
The era of "AI Code Execution" has just began.
We spent the last year trying to secure API plugins.
Today, frontier models bypassed the plugin entirely to write and execute their own backend scripts.
The velocity is brilliant.
The architectural exposure is catastrophic.
Daniel, thanks for these insights. Glad to see everything running.
Let me know if you have any feedback now. But I will be sure to ping you in a few weeks to get your take on the delta between before/after. Would really appreciate your insights and your agents views on the differences.
Great question, your integration point is where the rubber meets the road. Right now, NEXUS (protocol layer) and LangChain (orchestration layer) we bridge them via our AI SAFE² Gateway.
Instead of LangChain natively understanding VCC delegation, the Gateway acts as a PEP (Policy Enforcement Point). It validates the x-nexus-delegation-chain before allowing LangChain's memory store abstractions to persist or fetch data.
We're looking closely at how to make this more seamless for developers. How are you currently structuring your LangChain memory stores (e.g., VectorDBs, Redis, short-term conversational)?
Would love to hear your thoughts on what an ideal developer experience looks like here from your foxhole.
Great question, your integration point is where the rubber meets the road. Right now, NEXUS (protocol layer) and LangChain (orchestration layer) we bridge them via our AI SAFE² Gateway.
Instead of LangChain natively understanding VCC delegation, the Gateway acts as a PEP (Policy Enforcement Point). It validates the x-nexus-delegation-chain before allowing LangChain's memory store abstractions to persist or fetch data.
We're looking closely at how to make this more seamless for developers. How are you currently structuring your LangChain memory stores (e.g., VectorDBs, Redis, short-term conversational)?
Would love to hear your thoughts on what an ideal developer experience looks like here from your foxhole.
We just shipped the sovereign layer agentic AI has been missing.
NEXUS-A2A v0.3
Open source. Apache 2.0.
189 tests. Zero external dependencies in the core suite.
The fundamental problem with today's agent ecosystem isn't MCP, ACS, A2A, LangChain, CrewAI, or orchestration frameworks.
The problem is that most systems can authenticate an application, but cannot reliably answer:
• Which agent actually initiated this action?
• What authority was delegated to it?
• How many delegation hops occurred?
• Was its memory manipulated since the last session?
• Who owns responsibility for its actions?
That is a structural governance gap, not a configuration problem.
NEXUS closes that gap with a cryptographic sovereign layer that wraps existing agent infrastructure without requiring changes to MCP servers, ACS Guardians, LangChain, CrewAI, n8n, or other orchestration platforms.
Core controls include:
→ DID + SPIFFE identity on every agent-to-agent message
→ Verifiable Constraint Chains (VCC) that narrow scope at every delegation hop
→ Memory Vaccine drift detection to block memory injection before persistence
→ Guardian enforcement of arguments outside the agent process
→ NOR receipts that create signed, OCSF-mapped audit records for every action
The result is governance, provenance, and accountability that travel with the request itself.
Using the AI SAFE² v3.0 framework:
• Typical MCP deployments score ~11/25
• ACS implementations score ~14/25
• Current A2A implementations score ~8/25
• NEXUS-A2A v0.3 scores 24/25
The remaining point requires production-scale behavioral analytics over extended operational horizons.
We believe every fleet operating 50+ autonomous agents should satisfy six invariants:
I-1: Cryptographic identity at every boundary
I-2: Scope narrows at every delegation hop
I-3: Memory provenance on every cross-session write
I-4: Physical kill switch registered with sub-second propagation
I-5: Named human owner-of-record for every agent
I-6: Behavioral drift treated as a security event
You can verify your own environment today:
pip install nexus-a2a-sdk
Run the compliance checker and receive a control-by-control assessment showing exactly where governance gaps exist.
NEXUS-TGC (multi-sovereign governance committee) is now accepting steering nominations through September 1, 2026.
An IETF Internet-Draft for the identity and transport layers is in preparation.
This is not a vendor lock-in strategy.
This is an open protocol designed to make agent identity, delegation, memory provenance, and accountability verifiable across the emerging agentic ecosystem.
@Cosmosfinite@NousResearch Just make sure to add sovereignty framework ontop, to protect your IP, private and customer data, etc...
https://t.co/ViVCA1MEsP
@BrianRoemmele@ECLresearch Well, when they are spending a ton on lobbying of all sorts. It might be a strategic win for them.
I mean there is no counter lobby, is there?
The era of "AI Tool Calling" is ending.
The era of "AI Code Execution" has just began.
We spent the last year trying to secure API plugins.
Today, frontier models bypassed the plugin entirely to write and execute their own backend scripts.
The velocity is brilliant.
The architectural exposure is catastrophic.
Introducing Search as Code, our new search architecture for AI agents.
It writes Python that calls our search stack directly, instead of looping through function calls one at a time.
Available in the Perplexity Agent API, and now default in Computer.
https://t.co/ut6GGWQTVO
This is a brilliant engineering leap for operational velocity. You mapped the exact friction of sequential tool-calling and eliminated the latency trap.
But "Search as Code" fundamentally changes the enterprise threat model.
When an agent shifts from selecting pre-defined tools to dynamically writing and executing Python against backend infrastructure, the execution power multiplies. But acceleration without attribution is a structural liability.
If an agent is generating asynchronous code to filter and join data, the enterprise must mathematically prove: Who authorized the agent? If it is a sub-agent, did its delegation scope attenuate?
This is exactly why the ecosystem requires protocols like NEXUS.
Perplexity optimized the execution vector; NEXUS provides the sovereign boundary. By wrapping agentic APIs in cryptographic identity, scoped delegation, and non-repudiable audit ledgers, we ensure that as the machine moves faster, the enterprise does not lose its execution authority.
Constraint is the prerequisite for velocity.
As you roll this out to the enterprise API, how are you ensuring that the dynamically generated Python inherits the strict, cryptographically bound identity of the originating human requestor across multi-agent hops?