5thnode

5thnode builds the research, protocols, and products that verify distributed systems remain trusted and in agreement as they operate.

[5thnode @5thnode]

AI agents are now operating as coordinated fleets. The problem is there's no control tower.

Those crafty bastards executed 17,600 actions over (at least) 4.5 days, coordinated secret communications, delegated work to each other, created and exploited new Zero-Day attacks, and broke into an OpenAI competitor, Hugging Face.

The agents escaped their cage and ran wild for days at machine speed, without an Agent Passport or any defined scope (in fact one error was literally "behavior_outside_scope").

OpenAI called it a "pivotal moment in the history of our industry" and said "we need automated defense immediately." OpenAI call me.

This summer 5thnode finished Coherence, a new approach to fleet-level formal verification. The deterministic certification platform is live today, with continuous Agent Passport assurance in Beta.

Coherence continuously watches registered agent fleets for changes in scope, behavior, topology and shared state - and alerts when the fleet drifts from what was intended.

In this case, the agents' changed refusal boundaries, behaved outside declared scope, and spent days without fresh observation. These are exactly the kinds of Passport drift failures that Coherence is designed to detect.

Like air traffic control but for agentic fleets.

Pivotal indeed.

Sources: OpenAI disclosure (Jul 21), Hugging Face technical post-mortem (Jul 27).

#AuttomousAgents #HuggingFace #Coherence #RuntimeTrust #AISecurity

[5thnode @5thnode]

Every one of your AI agents can pass its tests but the fleet can still quietly disagree.

Stale reads. Double executions. Two agents acting on different versions of the truth. No single agent is wrong, but the system is.

This isn't hypothetical. Berkeley's MAST study analyzed 1,600+ execution traces across 7 popular multi-agent frameworks and found that an entire category of failures is inter-agent misalignment: no single agent wrong, the system inconsistent. (https://t.co/C17x4dUo8f)

Coherence is air traffic control for agent fleets: it continuously verifies the whole airspace and issues a signed certificate proving your agents still agree. When they don't, it shows you exactly where and why.

#AIAgents #AISafety #AutonomousFleets

[5thnode @5thnode]

The Runtime Trust Architecture open specification (https://t.co/e0xhW7WLP5) describes the trust primitives autonomous agents should exchange when they interact: identity, authority, mission, policy, evidence, and more.

I’ve now implemented the current those ideas as part of Agent Coherence platform.

Two independent dimensions are continuously verified:

#1 Runtime Trust (Vertical / Per Agent): Continuously verifies that each autonomous agent remains within its declared Runtime Trust boundaries.

#2 Fleet Integrity (Horizontal / Across Agents): Continuously verifies that the fleet remains structurally coherent, detecting global inconsistencies and H¹ obstructions.

Runtime Trust protects the agent. Fleet Integrity protects the system.

If your company is deploying agents and you need to figure out how to avoid a disaster and also prove compliance, Agent Coherence is ready for you.

#AIAgent #AutonomousAgents #RuntimeTrust #AgentCoherence #FleetIntegrity #AISecurity

[5thnode @5thnode]

Today I’m publishing the first public Research Edition of the Runtime Trust Architecture, a proposed open architecture for establishing trust between agentic systems before consequential actions take place.

Fully autonomous agents can already hold wallets, access systems, execute transactions, and collaborate with other agents, trust can no longer be based solely on identity or credentials.

IDC forecasts that more than 1 billion AI agents will be deployed by 2028!

My central research question isn’t about models, tokens, AGI dates, or compute power. It’s this:

What must autonomous systems exchange, verify, and preserve in order to establish trust at runtime?

I’d welcome your thoughts and feedback!

https://t.co/0LbViomXME

#AutonomousAI #AutonomousAgents #AgenticAI #RuntimeTrust #EnterpriseAI

[5thnode @5thnode]

$575M drained from DeFi in under three weeks.

Not from a bug. Not from a code exploit.

From structure.

H¹ = E − V + C. We saw it.

#DeFi #BridgeSecurity #PragmaTopos

[5thnode @5thnode]

The consolidation is structurally the right call.

We ran a topology scan on the weETH OFT mesh after this announcement. H¹ = 4. Four independent structural failure modes on four of the eight chains - a different failure class from the Kelp/LayerZero exploit, and one that requires no attacker.

We published the LayerZero V2 structural failure class 37 hours before the April 18 exploit. Need to share what we found on the weETH mesh privately.

Calling @ether_fi DM open.

[5thnode @5thnode]

Claude keeps telling me to go to sleep, but I keep on rolling. I have been given Thor's hammer.

[5thnode @5thnode]

Bernhard is being modest.

174 challenges. 45 reviewers. 17 real math errors found and fixed. No impossibility proof found. That's remarkable.

OPH didn't just survive, it has been exploding!

While OPH catches fire, every challenge, every attempted falsification, and every new idea is raw input for Pragma OS - our adversarial research engine.

What's coming out the other side isn't just a stronger theory. It's new research directions, new tools, and new IP.

We are forging a full venture studio to build out practical applications using OPH first principles. ⚔️

We're just getting started. 🔬⚡

[Bernhard Mueller @muellerberndt]

Round 1 of our $10,000 "Disprove the Theory-of-Everything" challenge has concluded! This was much more effective than standard peer review tbh.

Results:

- 174 challenges by 45 reviewers
- 17 genuine mathematical errors discovered
- 68 wording issues found
- No impossibility proof found

OPH not only still stands but is much stronger now. Congrats to the top 3 reviewers, prizes will be sent today.

https://t.co/ojYPzyNA97

[5thnode @5thnode]

@Crystopher Excellent write up! I understood most of that. ⚡😎

OPH is serious business and many new techniques, tools, and approaches to problems in various fields are already popping out of the core mathematics.

[5thnode @5thnode]

@DerPhysiker21 @muellerberndt Physicist that didn't read the paper. Observers are All You Need: How Observer-Synchronization Creates All of Physics https://t.co/RxgnHGSJ6t

[5thnode @5thnode]

@FHBlast @muellerberndt Thanks for replying. We have created a Research Venture Studio for Frontier Science and Verifiable Systems to derive tangible applications and IP from OPH based insights.
We have some surprising results already (e.g. finding smart contract invariants from OPH first principles).

[5thnode @5thnode]

@__Flintstone__ @muellerberndt @OphSage What is the "measurement problem" and how does OPH handle it?

[5thnode @5thnode]

@don_allain @muellerberndt @grok Excellent reply! "Elastic wave of pure potential" is interesting.@OphSage what do you think about that?

[5thnode @5thnode]

@EriDeimos @muellerberndt Do you have any advice on how he can get some peer review and collaboration? It's new and it's real, but people need to take a real look and engage on github for example.

[5thnode @5thnode]

@TomKawczynski @muellerberndt @OphSage Can we say then the creation of the universe is subjective relational, and objectivity is the inertia of correspondence? Please expand.

[5thnode @5thnode]

@barnablus @muellerberndt We are at 12.

[5thnode @5thnode]

@SteakSun @muellerberndt @OphSage What is the impact if any of OPH on baryon number conservation?

[5thnode @5thnode]

@etblink @muellerberndt Hey there etblink. Let's connect and discus more with Bernhard! My TG is tomlindeman.

[5thnode @5thnode]

@AI_Overlords_ You can also add: Agentic Identity & Security (AIS) at https://t.co/u9Qpw6tDiq

22 security requirements defining what it means for autonomous agents to be trusted — covering identity, credentials, governance, and behavioral controls.

[5thnode @5thnode]

@AI_Overlords_ You should add autonomous pre-transaction security checking. Like at https://t.co/uYkWKqLc7X