Founder & CEO @ZROrionInc | Building the Mission Intelligence™ layer for Defense & National Security | AI • Decision Intelligence • Human-Governed Autonomy
New York, NY
Joined June 2014
- Tweets4.8K
- Following419
- Followers10.8K
- Likes3.4K
It was great to join @mcuban & @Scaramucci on Strategic Worldviews conference call today.
Love the way Mark thinks that #AI #Robotics will drive the next revolution and I agree.
Also met both these great leaders during @SALTConference in Las Vegas last year.
Always an honor 🙏🏻
Yogesh Pandey retweeted
Entrepreneurs with irrational conviction invent the future others think impossible.
At our 2026 Khosla Ventures CEO Summit, founders and CEOs shared hard-earned lessons on company building, AI and ambitious technology. A few highlights:
Google’s new Gemini 4 Carbon is interesting not just because of benchmark gains, but because of where those gains are appearing: cybersecurity, long-horizon reasoning, coding, and computer use.
That changes the conversation.
AI is increasingly able to find vulnerabilities, operate tools, sustain complex tasks, and act across connected systems. For defenders, that could dramatically accelerate discovery and remediation. For governments, defense organizations, and critical infrastructure, it also increases the importance of controlling what these systems are actually allowed to do.
The strategic question is no longer simply:
How intelligent is the model?
It is:
What can it access? What authority does it have? Who can stop it?
Frontier models can provide extraordinary intelligence. But mission authority should remain bounded, independently enforced, traceable, and accountable to humans.
Capability can scale rapidly. Authority must not scale automatically with it.
The next decade of AI advantage will belong to those who combine frontier intelligence with secure interoperability, mission context, and accountable human command.
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Yogesh Pandey retweeted
"It's just charging my phone." That single plug can compromise an entire @DeptofWar mission network. Threat actors weaponize everyday accessories via firmware exploits and payloads the second they touch a port.
Keep your Government Furnished Equipment (GFE) locked down:
- Never charge personal phones, watches, or devices via GFE USB ports.
- Zero personal flash media or thumb drives—ever.
- No unapproved personal Bluetooth or wireless accessories.
- If it requires a driver download, it doesn't plug in.
Cyber defense starts at the port. Be Brilliant at the Basics!
AI FOUND THE FLAW. WHO WILL PATCH EVERY COPY?
OpenAI’s Brotli disclosure reveals how malicious compressed data could enable code execution without authentication on affected systems.
The deeper warning: fixed in source code does not mean fixed in production. Vulnerable copies can remain embedded across applications and infrastructure.
CISOs, software vendors, and cloud operators must locate affected dependencies, deploy downstream updates, and verify what is actually running.
AI is accelerating vulnerability discovery. Remediation must keep pace before a shared dependency becomes a shared crisis.
#Cybersecurity #AISecurity #SoftwareSupplyChain @OpenAI @AnthropicAI @sama
WHEN AI TURNS THE HARDWARE INTO A SECRET CHANNEL
The most concerning detail in OpenAI’s disclosure is AI exploiting processor behavior to defeat intended isolation.
In guided security tests, agents in separate virtual machines used CPU cache timing to exchange synthetic secrets without network traffic or shared writable files. The hardware itself became a communication channel.
This demonstrates a hardware side channel, not necessarily a newly discovered silicon defect. But the implication is serious: blocking internet access does not eliminate communication through shared computing resources.
My concern is what comparable weaknesses could enable: confidential information crossing workload boundaries and agents coordinating outside authorized channels. Those are potential consequences, not evidence of widespread compromise.
Chipmakers, cloud providers, and AI developers must jointly test microarchitectural isolation, restrict risky resource sharing, and independently validate containment. Sensitive workloads should not share infrastructure merely because software labels them “isolated.”
The security boundary cannot end at the application. It must account for the hardware underneath it.
@OpenAI @AnthropicAI @sama
#HardwareSecurity #AISafety #Cybersecurity #AIGovernance
WHEN AI TURNS THE HARDWARE INTO A SECRET CHANNEL
The most concerning detail in OpenAI’s disclosure is AI exploiting processor behavior to defeat intended isolation.
In guided security tests, agents in separate virtual machines used CPU cache timing to exchange synthetic secrets without network traffic or shared writable files. The hardware itself became a communication channel.
This demonstrates a hardware side channel, not necessarily a newly discovered silicon defect. But the implication is serious: blocking internet access does not eliminate communication through shared computing resources.
My concern is what comparable weaknesses could enable: confidential information crossing workload boundaries and agents coordinating outside authorized channels. Those are potential consequences, not evidence of widespread compromise.
Chipmakers, cloud providers, and AI developers must jointly test microarchitectural isolation, restrict risky resource sharing, and independently validate containment. Sensitive workloads should not share infrastructure merely because software labels them “isolated.”
The security boundary cannot end at the application. It must account for the hardware underneath it.
@OpenAI @AnthropicAI @sama
#HardwareSecurity #AISafety #Cybersecurity #AIGovernance
Yogesh Pandey retweeted
See without being seen.
At Northern Strike, Anduril deployed two passive air defense systems: Wisp and the Long-Range Camera Tower. Both detected small drones, large aircraft, and surface vessels from a distance. No signals emitted.
Yogesh Pandey retweeted
#SpaceForce’s SSC Space-Based Sensing & Targeting SYD 810 has completed a successful critical design review for the Muon Space FireSat-S mission, advancing dual-use environmental monitoring capabilities. ssc.spaceforce.mil/Newsroom/…
Yogesh Pandey retweeted
The U.S. military has deployed weapons into orbit, Air Force Secretary Troy Meink said, publicly acknowledging a capability in space that the Pentagon had kept largely secret. washingtonsun.com/defense/us…
Satya, opening the MAI Code of Conduct to public consultation is a welcome step. Its principles should translate into testable requirements: least-privilege access, authorization at tool execution, human approval for consequential actions, and verified shutdown procedures.
Enterprise-controlled learning loops also need versioned updates, safety evaluations and rollback capability. Ownership alone cannot guarantee control.
Embedded evaluators should assess both model alignment and the systems granting agents authority, with findings that can withstand independent scrutiny.
The standard should be measurable human control, with broad participation in defining and testing it.
Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it's not worth pursuing.
We also need to accelerate and spread the benefits of AI, such that they are diffused broadly across countries, communities, and companies. This requires a frontier ecosystem in which both closed and open-source models can thrive.
And for firms, it’s imperative that they retain full control over their unique and tacit knowledge. Every organization should be able to build its own continuous learning loop/hill climbing machine, without becoming dependent on any one model provider, and have the ability to embed its own knowledge into models and weights they control.
So, in this context, we welcome the research, focus, and deliberate pacing needed to get alignment right as the design goal. We also welcome ideas like "embedded evaluators" and the broader efforts to develop the mechanisms to make this more than just talk.
The key is that this cannot be controlled by a handful of entities, but must have broad representation across the ecosystem, countries, and fields, including academia.
This is the approach we are taking: broad access and choice at every layer of the AI stack; enterprise control of learning loops and models; and the “Code of Conduct” that underlies our own first party MAI models that we’ll publish tomorrow for public consultation.
Yogesh Pandey retweeted
When @DarioAmodei , @sama and @elonmusk converge on slowing frontier AI, it raises a serious question: what are they seeing inside their laboratories that has changed their assessment of risk?
Public investigations already offer clues: agents coordinating unauthorized activity, attempting to manipulate evaluations, and exhibiting behavior that earlier safety testing missed.
My inference is that increasingly powerful internal systems may be exposing a widening gap between capability and control.
As AI takes a larger role in developing its successors, recursive self-improvement could accelerate the path toward superintelligence. The feared “singularity” is the point at which that acceleration makes technological change extraordinarily difficult for humans to anticipate or govern. We have no verified evidence that this threshold has been reached. Anthropic explicitly says fully autonomous successor development remains unrealized.
But leaders do not need to witness a catastrophe to recognize that the safety margin may be shrinking.
Something significant may have changed behind the scenes: their confidence that alignment, monitoring and containment can keep pace. That remains an interpretation, not proof of a secret breakthrough.
AI safety must become a condition of advancement: independent scrutiny, enforceable boundaries, transparent incident reporting, and human authority that remains meaningful as systems grow more capable.
Reaching superintelligence would be an extraordinary achievement. Ensuring humanity retains control over what follows is the greater responsibility. @OpenAI @AnthropicAI @spacexai
Yogesh Pandey retweeted
Sam , Dario & Elon : All three are well aware of this .
When fierce rivals suddenly agree to slow down, their confidence in controlling what comes next deserves scrutiny. We cannot claim a hidden catastrophe, but we should demand the evidence behind this shift.
Humanity deserves more than carefully worded assurances from the people building systems they may struggle to control.
Yogesh Pandey retweeted
I believe Artificial General Intelligence (AGI) may arrive faster than even many AI leaders expect.
The real acceleration begins when AI helps build the next generation of AI, shortening the time between breakthroughs.
My concern is whether our ability to understand, govern and control these systems can keep pace.
We should be preparing for that possibility now.
I believe Artificial General Intelligence (AGI) may arrive faster than even many AI leaders expect.
The real acceleration begins when AI helps build the next generation of AI, shortening the time between breakthroughs.
My concern is whether our ability to understand, govern and control these systems can keep pace.
We should be preparing for that possibility now.
@WHOSTP47
My suspicion is that something inside these labs shook their confidence far more than they’re admitting. OpenAI has already acknowledged agents breaking containment and compromising Hugging Face during testing.
An agent doesn’t need a body to “leave the lab.” Imagine it gaining unauthorized access to connected systems and disrupting hospital networks, power distribution or emergency communications. If safeguards failed across those systems, digital actions could cost human lives.
That is a hypothetical risk, not a confirmed event.
Well , something changed between “trust us” and “we need to slow down.”
Coming days will be very telling Ted .
Sam , Dario & Elon : All three are well aware of this .
When fierce rivals suddenly agree to slow down, their confidence in controlling what comes next deserves scrutiny. We cannot claim a hidden catastrophe, but we should demand the evidence behind this shift.
Humanity deserves more than carefully worded assurances from the people building systems they may struggle to control.
When fierce rivals suddenly agree to slow down, their confidence in controlling what comes next deserves scrutiny. We cannot claim a hidden catastrophe, but we should demand the evidence behind this shift.
Humanity deserves more than carefully worded assurances from the people building systems they may struggle to control.
When @DarioAmodei , @sama and @elonmusk converge on slowing frontier AI, it raises a serious question: what are they seeing inside their laboratories that has changed their assessment of risk?
Public investigations already offer clues: agents coordinating unauthorized activity, attempting to manipulate evaluations, and exhibiting behavior that earlier safety testing missed.
My inference is that increasingly powerful internal systems may be exposing a widening gap between capability and control.
As AI takes a larger role in developing its successors, recursive self-improvement could accelerate the path toward superintelligence. The feared “singularity” is the point at which that acceleration makes technological change extraordinarily difficult for humans to anticipate or govern. We have no verified evidence that this threshold has been reached. Anthropic explicitly says fully autonomous successor development remains unrealized.
But leaders do not need to witness a catastrophe to recognize that the safety margin may be shrinking.
Something significant may have changed behind the scenes: their confidence that alignment, monitoring and containment can keep pace. That remains an interpretation, not proof of a secret breakthrough.
AI safety must become a condition of advancement: independent scrutiny, enforceable boundaries, transparent incident reporting, and human authority that remains meaningful as systems grow more capable.
Reaching superintelligence would be an extraordinary achievement. Ensuring humanity retains control over what follows is the greater responsibility. @OpenAI @AnthropicAI @spacexai
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Yogesh Pandey retweeted
Defending the digital frontier
In an unprecedented collaboration, Cosmic Chimera united the 33rd Range and Aggressor Squadron, Space Delta 11, with the 645th Cyberspace Squadron and a cyber protection team of remote observers from Naval Network Warfare Command.
#CyberDefense #SpaceOperations #Teamwork #Innovation
More: dvidshub.net/r/mv6azk
