Yaser retweeted
To our Blackbit community:
We want to make it easier for you to get started. That’s why we’ve lowered our minimum spend from $100 to $50.
If you signed up but haven’t taken the next step, we’d love to welcome you back →
blackbit.sh/?utm_source=orga…
Yaser retweeted
We’ve set up a test account with $50 in credits so you can try BlackBit VOID gateway models
Sign in at blackbit.sh
Account ID: 3830 7115 5997 2127 9128
Yaser retweeted
ORGN: Why TEE Matters for AI Coding
Encrypting code while it is stored is important.
Encrypting it while it is being sent over the internet is also important.
But there is still one moment where sensitive data can become exposed: while it is being processed.
This is where Trusted Execution Environments, or TEE, become important.
When an AI coding agent works on a repository, it may need access to source code, configuration files, internal documentation, credentials, or other sensitive information.
In a traditional cloud environment, that workload still depends heavily on trusting the infrastructure provider.
ORGN takes a different approach.
Its confidential environments are designed around hardware-based isolation, including Intel TDX, which creates protected virtual machines where workloads can run separately from the underlying host infrastructure.
The idea is simple:
Your code should remain protected even while the AI is actively working with it.
This matters because encryption at rest and encryption in transit only protect data before and after computation.
TEE focuses on the part in the middle.
The execution itself.
For developers, that means an AI agent can analyze code, run commands, modify files, or work inside a development environment without treating the entire cloud infrastructure as a trusted party.
But ORGN goes one step further.
TEE alone still leaves an important question:
How do you know the workload actually ran inside the protected environment you were promised?
That is where attestation comes in.
Hardware attestation provides cryptographic evidence about the environment where a workload was executed.
Instead of relying only on a provider saying:
“Your code was protected.”
Users can verify evidence tied to the actual execution environment.
This is especially important for companies working with proprietary code, regulated data, internal infrastructure, or systems where exposure could create serious security risks.
As AI agents gain more access to real production systems, the security model around them becomes just as important as the model itself.
The future of AI coding will not only be about how capable the agent is.
It will also be about where it runs, what it can access, and whether that environment can be verified.
That is the role TEE plays inside ORGN. @orgn_official
Yaser retweeted
We’re heading to SF for @stepconference.
AI is moving fast. Sensitive data and code should not have to move with it.
If you’re navigating privacy, security, or regulated data in AI, let’s talk 🤝
Yaser retweeted
If Mythos and Fable are sensitive enough to be export‑controlled and used by CISA to audit gov code, but a small team like ours can beat it on offensive cyber benchmarks in weeks at a fraction of the cost…
Maybe “frontier AI” is more narrative than scarcity.
We built a cyber-capable AI model that, based on benchmarking and real-world testing, is perhaps the most capable offensive-security LLM we have seen
It scores 95% on Cybench, solving 38/40 tasks
But benchmarks only go so far, so we pointed it at real systems
Through authorized programs, we tested it against a Fortune-500 technology company, a major global telecom, and leading Web3 protocols.
The result: a 100% success rate - it found a viable path into every authorized target in our test set.
It adapts, reasons, and finds routes that aren’t obvious, even in systems that already passed audits and security reviews.
If this holds at scale, most current security assumptions break.
Yaser retweeted
We built a cyber-capable AI model that, based on benchmarking and real-world testing, is perhaps the most capable offensive-security LLM we have seen
It scores 95% on Cybench, solving 38/40 tasks
But benchmarks only go so far, so we pointed it at real systems
Through authorized programs, we tested it against a Fortune-500 technology company, a major global telecom, and leading Web3 protocols.
The result: a 100% success rate - it found a viable path into every authorized target in our test set.
It adapts, reasons, and finds routes that aren’t obvious, even in systems that already passed audits and security reviews.
If this holds at scale, most current security assumptions break.
Love this
Private equity generally aims to buy businesses, cut costs, and flip them in 4-7 years for a profit.
As an alternative, we started a company to buy and hold businesses, and back them up with a bitcoin treasury:
lynalden.com/orange-juice/
Yaser retweeted
Not all private AI is the same, so we make that distinction very clear.
When choosing your AI model in the IDE, you will see tags: ZDR or TEE
ZDR = Policy. The provider contractually commits to not retaining or train on your data.
TEE = Silicon. The model is deployed on Intel TDX & NVIDIA TEEs. Privacy is enforced by the hardware itself.
Non-US devs: Tired of US frontier models locking you out or hoovering up your IP?
GLM-5.2 is now LIVE on ORGN, frontier-level coding & agentic power (1M context, crushes frontend/agent workflows, close to Fable on real builds) with zero default data retention + confidential chips + hardware attestations.
Your code. Your IP. No compromises.
AI sovereignty will define national power for the next generation.
Watch how the US sells fighter jets, and you already know how it will sell advanced AI.
After years building intelligence systems inside the defense world, none of this surprises me. No ally flies what the US flies. They get an export model: capability capped, the crown jewels withheld.
Frontier AI just started down that road. This week Anthropic was ordered by the US government to cut its most capable models off from every foreign national, even staff inside the US. Lesser models were untouched. That is the first export-tiering of AI, and not the last.
So for an ally the question is not whether you can buy advanced American AI, but which grade. The top tier stays home. Real capability cannot be bought. It has to be built.
Here is what most analysis misses: the high ground keeps moving up the stack. First chips, then models. The next contest, the one that decides who is sovereign, is the agentic layer: the systems that turn models into agents acting in your most sensitive environments. The model becomes a component. The layer that governs it is the prize, and the one layer a nation can own outright.
The roadmap, in order: compute on your own soil, an open-weight foundation you control, models trained on your own data and language, and the agentic control layer on top, governing which agents run, what they reach, and proving every action. Foreign frontier systems stay at the margins, never under what must keep running.
Every nation faces one choice: build the layer that acts, or rent it from someone who can take it back. Those who build early set the terms.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
