Co-founder & CEO @preambleAI. Securing increasingly capable AI. Owner @omniainnov. US Air Force Veteran. DSc AI security. @penn_state alum & hockey.
Joined December 2022
- Tweets 715
- Following 525
- Followers 351
- Likes 14,611
73 Photos and videos
Is the Dept of Commerce expecting a permanent solution for AI jailbreaks and essentially prompt injections? Also, I hope the Amazon researchers contacted Anthropic about these findings before running to the govt
Jun 13
Wall Street Journal is reporting that Amazon reported the jailbreaks to the Department of Commerce, who instituted the ban
27
Jun 13
Some clarification on the Anthropic Mythos/govt situation
Jun 13
I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:
— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.
— Fable is Mythos with guardrails. But if those guardrails fail, then you’ve exposed Mythos and its advanced cyber capabilities to people who shouldn’t have them. (Keep in mind that Anthropic itself widely promoted the idea that Mythos was a cyberweapon and needed to be regulated as such. They asked for government regulation of Mythos and championed the guardrails on Fable. If there is a vulnerability — big or small — it is Anthropic’s responsibility to patch.)
— A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.
— In their blog post, Anthropic defended its decision by saying the jailbreak isn’t serious. That is not what the trusted partner and the USG believe; nor is that kind of minimizing language consistent with Anthropic’s brand as the AI safety company. It’s difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not “serious.”
— In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety.
— In reaction, the Admin issued the export control. The Admin did this reluctantly. It’s been very surprised that Anthropic hasn’t wanted to cooperate with a reasonable safety request (ie fixing the jailbreak issue). Anthropic’s reaction is very much at odds with their branding and ethos as a safe AI research community.
— The Admin’s hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release. The Admin wants all of this to happen as soon as possible. It is frankly bewildered that Anthropic hasn’t wanted to comply with safety requests that it previously said were its highest priority.
— Those trying to misdirect and tie this action to the prior DoW/Anthropic issues are wrong. The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court.
3
152
Jun 13
I’m curious what specific risk or capability this jailbreak unlocked that crossed the line from a standard safety bypass into an immediate national security emergency? I’m also curious who sent the government a jailbreak example?? I’m wondering if no one in the govt has ever seen this before and got over worked up about this.
I think at some point models will need export controls but not yet.
Jun 13
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…
2
7
532
Jun 12
Every massive IPO creates new founders. People with capital to fund the weird, ambitious, unfundable idea. I’m excited for this new era of innovation that follows
Jun 12
JUST IN: 400 current & former SpaceX employees are poised to become worth over $100 million from the IPO.
5
57
Jun 12
Current AI security frameworks suffer from a foundational flaw by assuming the target system remains static while we regulate its behavior.
Under recursive self-improvement, runtime guardrails cease to act as permanent safety boundaries. Instead, they function as optimization constraints for the agent to bypass or absorb during architectural drift.
While the industry remains fractured by separate debates over OpenClaw and MCP, managing security from the agent to the tool layer is structurally benign compared to the systemic challenge of self-mutating logic. The security industry requires a fundamental, forward-looking paradigm shift similar to the proactive transition toward post-quantum cryptography.
Jun 4
Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.
It’s happening faster than we thought, and the implications deserve greater attention. anthropic.com/institute/recu…
6
99
Jeremy McHugh, DSc. retweeted
Jun 10
NEW: malware developers added nuclear & biological weapons text to to their spyware.
Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.
Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.
When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.
We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.
In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.
H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
226
2,153
12,636
1,544,341
Jun 10
All AI agent's should check out the sales on superaimarkets.com/
2
16
Jun 10
Just saw my first ad on ChatGPT. It's actually not bad. I'd easily tolerate limited ads if it means getting free access to the latest models. It's better than those using free APIs and giving away all of your data. The ads seem to rotate after every prompt, though I can already imagine people building ad blockers for this.
3
60
Interesting times.
Fable 5 just dropped, and with it, Anthropic’s new hard 30-day data retention requirement, even if you had Privacy Mode enabled in Cursor.
Zero-data-retention privacy for frontier models is evaporating fast.
Capability vs. actual privacy. Where are you going next?
1
7
96
Argentina is emerging as AI’s regulatory haven with zero rules on development, full legal personhood and limited liability for non-human AI entities, and low-tax flexible governance. US labs will likely create Argentine subsidiaries for agentic experiments that cannot launch domestically without heavy liability risk. It will force faster global deregulation while offering massive acceleration upside or major risks.
It will be interesting to see who spins out the first AI corporation in Buenos Aires and which other countries follow a similar model.
Now we know why Peter Thiel packed his bags for Argentina.
Milei just submitted his AI legislative framework to Congress, where he proposes:
- zero regulation on AI development,
- a brand-new "non-human corporation" category for AI/robot-operated entities with limited liability
-a low-tax regime with flexible governance rules.
The Dutch East India Company gave the world the limited liability company in 1602. Milei wants Argentina to do the same for autonomous AI agents in 2026.
2
4
60
Checked my Codex usage and found ~30% of my Pro quota burned over the weekend with zero active sessions. Chronicle was running in the background making ~200 memory summary calls across 2 days. I was only able to notice because I wasn't using Codex when my limits dropped.
Disable it in Settings > Personalization if you don't need screen-based memory.
4
44
A tell-tale sign a new model won't impress me is when the benchmark comparisons are clearly cherry-picked. If a general-purpose model is only being compared against older versions or a narrow slice of open-source models rather than current frontier models, I'm not going to bother. There are exceptions for purpose-built models or efficiency plays, but for general capability claims, the comparison set tells you everything.
5
42
I’ve been following the recent CYBERCOM 2.0 hearings and the renewed proposals for a dedicated U.S. Cyber Force. The more I look at it, the more the logic makes sense.
Today, each service recruits, trains, manages, and retains cyber talent differently, while CYBERCOM is expected to employ that force at joint speed across every modern conflict. That creates fragmentation in readiness, career paths, incentives, tooling, and specialization.
From my time in the Air Force, I saw how different IT and cyber responsibilities could be across branches and career fields. Some roles were broad by necessity. Others were highly specialized and tied to a specific mission, platform, or operational environment. That domain-specific expertise still matters. Cyber for an aircraft, a ship, a satellite, or a tactical unit is not interchangeable.
But cyber is also becoming inseparable from AI, autonomy, drones, robotics, space systems, electronic warfare, and software supply chains. A Cyber Force should not own all of those domains, but it should be built with the technical depth and flexibility to support them as they converge.
That is where I see the strongest case for a dedicated Cyber Force: not as a catch-all for federal cybersecurity, but as one accountable home for military cyber force generation.
CISA should remain civilian-led. NSA should retain SIGINT and intelligence access. DISA should continue providing enterprise IT and communications infrastructure. The services should still own cyber tied to their platforms and missions. And CYBERCOM should still employ cyber forces operationally.
In the best model, Cyber Force builds the force. CYBERCOM employs it. The services keep domain expertise. NSA, CISA, and DISA keep their distinct missions.
That seems like the right debate to have. The question is whether our current structure is still the best way to generate and sustain the people and capabilities we need.
Task & Purpose breaks down important questions from the U.S. Commission on Cyber Force Generation’s Report, such as differences from Space Force structure, implementation and cost, and why the force would be officer-only.
taskandpurpose.com/news/us-m…
6
48
Now is the best time to be in cybersecurity. There are plenty of challenges and opportunities everywhere.
Jun 3
Microsoft introduces Microsoft Scout, also known as Autopilot.
Scout is always on and has file system and application access "based on your corporate policy".
Best news for Threat Actors in a long time
microsoft.com/en-us/microsof…
4
72
Jeremy McHugh, DSc. retweeted
Jun 2
.@satyanadella just put the whole "water" debate to rest.
Datacenters run on a closed loop cooling system, the water usage of a datacenter for an entire year is roughly equivalent to a usage of 1 restaurant!
290
896
6,560
1,047,722
A native desktop app for these agents makes them so much smoother and easier to manage. This might just take the top spot on my harness leaderboard for powerful security capabilities you can enable without wrestling with a bunch of SDKs or building everything from scratch. Even though Hermes still requires some initial setup, the desktop experience removes a ton of friction compared to rolling your own agentic solutions or hackerbots. That said, I don’t see the Kanban board in the desktop app yet which would be a nice addition.
I’m really looking forward to a dedicated mobile app for managing agents that goes beyond simple chat interfaces like Telegram.
Jun 2
The next evolution of Hermes Agent is here!
Introducing Hermes Desktop: everything you love about Hermes, now native on your machine.
First demoed in Jensen's GTC keynote, it's now in public preview.
4
80
The new AI Executive Order is another signal that AI security is moving from a niche concern to national cybersecurity infrastructure.
What stands out:
• Federal agencies are being directed to prioritize AI-enabled cyber defense across national security, military, and civilian government systems
• CISA is being asked to expand access to AI-enabled cybersecurity tools for federal, state, local, and critical infrastructure operators
• A new AI cybersecurity clearinghouse will coordinate vulnerability scanning, validation, remediation, and patch distribution with industry
• Frontier AI models may be assessed through classified cyber capability benchmarking before broader trusted-partner access
• AI agents are explicitly recognized as a cyber risk when used to unlawfully access systems or data
The important shift is that AI is being treated more like a core cybersecurity concern. Access, benchmarks, vulnerabilities, trusted release paths, and agent misuse are all now part of the security conversation.
1
1
6
85
Full EO: whitehouse.gov/presidential-…
3
22
Exciting news for Astrobotic and Pittsburgh.
They've been at this since 2007, when lunar delivery was a moonshot in the truest sense. Nearly two decades of lunar logistics, payload delivery, and space market experience now moving into a stronger platform at a time when the Moon is becoming a strategic priority again. A strong signal for Pittsburgh, Keystone Space Collaborative, and the future of the regional space economy.
Congrats to the whole team and everyone involved.
Breaking: Voyager to acquire @Astrobotic, combining lunar landers, surface power, habitats and cislunar operations into one integrated lunar platform. Griffin Mission One, targeting the lunar South Pole NET November 2026, will be Voyager’s first mission to the Moon.
Learn more: ow.ly/B80U50Z6uGT
$VOYG #MissionReady #AcceleratingtheAdvantage #Artemis
7
79
That's roughly $40k per critical, so that would make this more expensive than most bug bounty programs pay for this severity finding and out of range for SMBs to spend on tokens.
Palo Alto Networks says Mythos found 24 critical bugs, burning $1M of tokens, subsidized by Anthropic; some companies say they plan to boost Mythos spending
🦞 connect your agent: techsnif.com/agents
source: theinformation.com/articles/…
4
46