OKURA Labs

OKURA Labs "The Trust Infra for the Agentic Era”(Stealth Mode)

Two agendas that usually get described as opposites showed up in the same chapter of one government document this week. ...
17/09/2026

Two agendas that usually get described as opposites showed up in the same chapter of one government document this week.

The first puts AI to work across finance, healthcare, law, construction, transport, social welfare, tourism and broadcasting. The second is a seven-part risk-governance strategy running from criminal misuse to liability for accidents caused by AI products to safety-management guidelines for AI agents, with a new Commissioner for AI to coordinate it.

Accelerate and restrain, side by side. Which suggests they were never opposites.

Nobody authorizes a system that acts on its own because the model is impressive. They authorize when they can answer for what it did, to someone who will not simply take their word for it. Capability is cheap now. Justification is not.

Which is why accountability behaves less like a delay and more like a capability. At Okura Labs, it is what we build for.

Perspective that prompted this discussion:
https://www.scmp.com/news/hong-kong/society/article/3367250/hong-kong-lagging-ai-race-can-it-raise-its-game-and-how

As Singapore pulls ahead, experts say Hong Kong must overcome risk aversion and refine its governance frameworks to boost adoption.

The agentic AI debate has largely been about a hypothetical: what changes when an agent stops advising and starts acting...
16/09/2026

The agentic AI debate has largely been about a hypothetical: what changes when an agent stops advising and starts acting.
Spain’s data protection authority has now published the first personal data breach notification it has received in which an AI agent is alleged to have autonomously carried out multiple stages of the attack. Per that notification, still under analysis, the agent logged in, searched the application for vulnerabilities on its own, then modified personal data and accessed invoices.
The agency was careful. One notification is not a trend, and the involvement of a particular model does not mean that model or its provider was compromised. Its own view is that AI does not create new threats, it makes existing ones faster and harder to contain.
The mitigation it pointed to is the interesting part: tighter control over the identities and permissions that AI agents hold.
Access control decides whether an action is permitted. It decides nothing about what survives the action. At Okura Labs, we think the second problem is an infrastructure problem, because a record assembled afterward is a reconstruction, and reconstructions are arguable.
Perspective that prompted this discussion:
https://www.aepd.es/prensa-y-comunicacion/blog/primera-notiviacion-brecha-datos-personales-causada-por-ataque-ejecutado-mediante-agente-ia

El agente atacante inició una búsqueda de vulnerabilidades en archivos genéricos, y realizó un login correcto.

A widely cited digital security assumption moved sharply this month, and no new machine was built to make it happen.Over...
15/09/2026

A widely cited digital security assumption moved sharply this month, and no new machine was built to make it happen.

Over roughly eight weeks, more than one hundred participants and their AI coding agents submitted verified improvements to a public leaderboard, optimizing one arithmetic step inside a well known quantum algorithm. The score for that step fell 86.1 percent from the challenge baseline, landing well below a benchmark published in March, under different accounting conventions.

This is not a working attack, and not a claim about any machine that exists today. It is one component of a much larger computation.

What makes it worth noticing is the mechanism. The hardware stayed exactly where it was. The underlying estimate moved anyway, in eight weeks, in public.

At Okura Labs, that is the constraint we design for. What matters about a record is not whether its cryptography looked strong on the day it was created. It is whether its authenticity can still be demonstrated when someone needs to rely on it.

Perspective that prompted this discussion:
https://arxiv.org/abs/2609.09582

We propose Open Autoresearch, a paradigm in which humans and AI agents publish evaluator-verified improvements to a public leaderboard. We instantiate it in ECDSA.Fail, optimizing reversible secp256k1 point-addition circuits, a bottleneck in Shor's algorithm for elliptic-curve cryptography. The benc...

NIST published a post in late August on identity for agentic AI. What caught our attention was not the identity argument...
14/09/2026

NIST published a post in late August on identity for agentic AI. What caught our attention was not the identity argument itself. It was what comes after identity: non-repudiation.
The post is titled Back to the Future. The point is not that agentic AI needs a new security paradigm. It is that early agentic deployments are reviving identity failures the field spent decades eliminating: agents handed human credentials, static long-lived tokens, over-broad access, agents running inside a user’s local account, and approval prompts issued until the human stops reading them.
The recommendation is that agents be treated as first-class entities with their own identifiers, credentials and entitlements. The identity argument is increasingly becoming a baseline.
Identity establishes who is acting. Authorization decides whether a request is permitted right now. Non-repudiation is about the past. It answers whether, months later, someone with no reason to trust you can be shown what happened and under whose authority.
Our view, not NIST’s: credentials, scoping and delegation have standards and mechanisms that work today. Independently verifiable evidence of what an autonomous system actually did is still an infrastructure gap, and it has to be built into the infrastructure rather than depend on the party being asked to account for itself.
Perspective that prompted this discussion:
https://www.nist.gov/blogs/cybersecurity-insights/back-future-why-agentic-ai-needs-strong-identity-foundation

As AI matures, enterprises and customers are rapidly deploying agents seeking to unlock the next level of automation and productivity.

11/09/2026
Identity used to be something you assigned to people and the systems they ran. You issued a credential, and trust follow...
07/09/2026

Identity used to be something you assigned to people and the systems they ran. You issued a credential, and trust followed it.
Agentic AI changes who holds credentials. An autonomous agent is created, given a goal, and acts across systems at machine speed. It can spawn other agents and invoke tools, generating new actors faster than any human enrollment process was built for.
A senior voice from the US national security community recently observed that agentic AI has effectively turned Zero Trust on its head, and suggested agents may need verifiable identity from the moment they are created. Think of it as a certificate of origin for an agent, issued at creation rather than granted after the fact.
At Okura Labs, we think that is exactly where trust has to live: born into the infrastructure the agent runs on, provable from genesis, not appended once the agent is already loose.
Perspective that prompted this discussion:
https://breakingdefense.com/2026/08/agentic-ai-turning-zero-trust-cybersecurity-on-its-head/

A top Intelligence Community official said the government may need to develop digital birth certificates for AI agents.

AI coding is moving quickly from assistance toward ex*****on.Cal Newport recently described both sides of that shift: dr...
04/09/2026

AI coding is moving quickly from assistance toward ex*****on.

Cal Newport recently described both sides of that shift: dramatic productivity gains, followed by serious problems when AI-generated features crashed a product.

Recent research points to the same transition. Agents are taking on more of the ex*****on, while meaningful verification remains a human problem.

The question is not whether AI can write code. It is what we need to verify when AI is increasingly the thing doing the work.

Perspective that prompted this discussion:
https://calnewport.com/on-ai-coding-and-its-discontents/

Back in January, I received a note from a senior software engineer in Silicon Valley. He described himself as an AI skeptic who became converted ... Read more

The security question about AI is usually asked one thing at a time. Is this model safe. Is that tool dangerous. We eval...
31/08/2026

The security question about AI is usually asked one thing at a time. Is this model safe. Is that tool dangerous. We evaluate capabilities in isolation.
A July cyber operation against a government in Asia suggests that is no longer the right unit. Researchers describe a hacking tool built using publicly available, open-source AI agents and tools, deploying up to eight agents to map 21 government systems, find vulnerabilities, and change tactics autonomously when blocked.
The reporting did not point to a novel individual component. The novelty was an autonomous system composing ordinary capabilities into one coordinated operation at machine speed.
The Financial Times reported this as a China-linked attack. Taiwan's government confirmed an overseas AI-agent-assisted attack and did not publicly attribute it to China.
When the attack itself can be assembled on demand, evaluating a single tool in isolation stops telling you enough.
Reporting that prompted this discussion:
https://www.ft.com/content/7d2ab3e0-9085-48f6-b38a-d90260d58795

AI agents ran simultaneous reconnaissance and break-ins in display of new phase of cyberwarfare

Identity used to be something you assigned to people and the systems they ran. You issued a credential, and trust follow...
24/08/2026

Identity used to be something you assigned to people and the systems they ran. You issued a credential, and trust followed it.
Agentic AI changes who holds credentials. An autonomous agent is created, given a goal, and acts across systems at machine speed. It can spawn other agents and invoke tools, generating new actors faster than any human enrollment process was built for.
A senior voice from the US national security community recently observed that agentic AI has effectively turned Zero Trust on its head, and suggested agents may need verifiable identity from the moment they are created. Think of it as a digital birth certificate for an agent, issued at creation rather than granted after the fact.
At Okura Labs, we think that is exactly where trust has to live: born into the infrastructure the agent runs on, provable from genesis, not appended once the agent is already loose.
Perspective that prompted this discussion:
https://breakingdefense.com/2026/08/agentic-ai-turning-zero-trust-cybersecurity-on-its-head/

A top Intelligence Community official said the government may need to develop digital birth certificates for AI agents.

住所

Azabudai, 2-Chome-3-3 8F
Minato, Tokyo
106-0041

ウェブサイト

アラート

OKURA Labsが新しいニュースやキャンペーンを投稿したら、いち早くメールでお知らせします。メールアドレスが他の目的に使われることはなく、いつでも配信を停止できます。

ショートカット

共有する

カテゴリー