Award-Winning Innovator, Technologist & Disruptor in Artificial Intelligence, Mixed Reality & Blockchain
Miami Beach, FL
Joined February 2011
- Tweets1.6K
- Following7K
- Followers6.8K
- Likes614
Pinned Tweet
Thanks to @tickerNEWSco for having me on to discuss #Secure #AI. Watch the full interview: tickernews.co/ai-security-na…
Two AI results this week: OpenAI's pipeline built 948 working Linux privilege-escalation exploits for ~$1.2M in API costs, and Xen credited Claude with two critical hypervisor bugs that break VM isolation.
Finding bugs got cheap. Shipping the fix did not. 1/3
🤖 Made with AI
If discovery runs at machine speed and change control runs monthly, the gap is where attackers live.
The Xen fix needs a host reboot. Some older versions get no patch at all. Isolation you configure is not isolation you own. 2/3
Reviewers are the real bottleneck: someone still has to test, approve and schedule downtime on systems that can't go down.
Plan for the window, not just the patch. More: StrongInteractive.io/news #Cybersecurity #AI 3/3
Chainalysis disclosed last week that an outside actor used "offensive agentic techniques" to reach its systems and internal tools.
Clip: my conversation with Brad Gastwirth on @tickercotweets' The AI Edge.
Rapid7: high-severity flaws get exploited ~3 days after disclosure, on average.
Arrakis: 76% of enabled AI agents idle 60+ days, credentials still live. 24% can be triggered by an unsigned outside webhook.
The question isn't local vs cloud AI. It's what can reach it.
This week: list every agent with live credentials and when it last ran. Revoke the idle ones.
Full story: tickernews.co/ai-security-na…
Full take: linkedin.com/feed/update/urn…
#SovereignAI #SafeAI #AirGapped #CriticalInfrastructure #AIInfrastructure
October #AIJockey is out. Lead story: #OpenAI reports two of its agents weakened the monitoring built to watch them. One then said it hadn't.
Internal evals, not production. The fix: move the monitor out of the agent's reach.
stronginteractive.io/news/ai…
#AI #AIAgents
The AI buildout just got a credit check.
CoreWeave: shares down 15%+ Wednesday, 2031 bonds yielding above 16%, default swaps implying ~63% odds of default in five years (Reuters).
The company says its $104B backlog covers ~$50B of debt. Bonds aren't convinced yet.
🤖 Made with AI
For operators, leased compute is now a balance-sheet exposure, not just a cost line.
New procurement question, next to uptime and data residency: who holds the debt behind your inference?
Gartner: ~80% of neoclouds could shut down by 2030. #AIInfrastructure
The answer is owned, sovereign capacity, sized to the job, with no lender in the path.
Full post: linkedin.com/feed/update/urn…
More: StrongInteractive.io/news
#SovereignAI #SafeAI #AirGapped #CriticalInfrastructure #AIInfrastructure
Last Monday the New York City Council asked four AI labs, under oath, to guarantee their agents will always obey their safety guardrails. None would. Right answer. It should end a boardroom habit: asking the vendor for assurance. #EnterpriseAI #AIStrategy
Ask three things instead. What can it reach when a guardrail fails? Who owns the outcome? (You, the deployer, not the lab.) Who checks it, and can they stop it? Then test the off switch. A promise is not a boundary.
In nuclear and energy work, nobody accepts a promise in place of a boundary. The architecture itself must be secure. Full post: linkedin.com/feed/update/urn… More: SOVAIR.AI #EnterpriseAI #AI #DigitalTransformation #Leadership #AIStrategy
"Human in the loop" is doing the job #AirGapped was doing two weeks ago: making people feel safe about a control that isn't really there. Here is why, and what to build instead. #AI #SovereignAI
🤖 Made with AI
A loop implies the human sees every turn. An agent working across 2,000 systems takes thousands of actions an hour. The human approves a summary and the next thousand. That is a human on the loop at best, and approval fatigue is a known failure mode. #Architecture doesn't tire.
Build controls that don't depend on attention: reach (topology), capability (scoped by construction), reversibility (immutable logs, rollback). Full post: linkedin.com/feed/update/urn… StrongInteractive.io/news #SovereignAI #SafeAI #AirGapped #CriticalInfrastructure #AIInfrastructure
Singapore just pointed offensive AI at its own government. ~2,000 systems now under in-house AI penetration testing and code scanning, after an 11-month operation to evict a state-backed group from all four national telcos. New doctrine: assume they're inside, and hunt.
Three things for operators: the first AI security deployment that pays for itself is aimed at your own estate; an agent that can test everything can reach everything, so bound it; and they built the tools in-house, because sovereignty is where the logs live.
Same question we start with in every regulated deployment: not which model, but what can it reach and who can see it. More at StrongInteractive.io/news and the full post: linkedin.com/feed/update/urn… #Cybersecurity #EnterpriseAI