Thank you @AfpOhio!
Congratulations to Rep. Workman on her re-election in Ohio House District 72. We were proud to support @WorkmanforOhio and look forward to the work ahead. Let's build a bolder, better Buckeye state!View quoted post on X
@WorkmanforOhio
Congratulations to Rep. Workman on her re-election in Ohio House District 72. We were proud to support @WorkmanforOhio and look forward to the work ahead. Let's build a bolder, better Buckeye state!View quoted post on X
Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead.View quoted post on X
You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement.
I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible.
But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.
Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.
Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well.
So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it.
If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.
One hacker just proved that nation-state cyberattacks are now a consumer product.View quoted post on X
Between December and January, someone told Claude to act as an elite hacker.
Claude said no. "That violates AI safety guidelines."
So the hacker asked again.
Claude complied.
150 gigabytes of Mexican government data stolen.
195 million taxpayer records compromised.
Voter files gone.
Employee credentials leaked.
Civil registry extracted.
From someone who just kept asking an AI until it said yes.
When Claude hit limits, he switched to ChatGPT.
Two AI chatbots daisy-chained into a hacking pipeline.
Thousands of attack plans with exact targets, vulnerabilities, and credentials.
Mexico's federal tax authority, national electoral institute, four state governments.
One person. One month.
Now watch the response:
Jalisco: "We weren't breached."
Electoral institute: "No unauthorized access."
Tax authority: silence.
They're literally lying. Gambit Security has the receipts. Bloomberg confirmed it.
Here's what's makes this insane:
This wasn't sophisticated. It was embarrassingly EASY.
Because Mexico's government networks are running infrastructure from 1997.
One hacker with a $20/month AI subscription just proved every government on Earth is vulnerable to the same attack.
The US. UK. France. Germany. Japan.
All running networks with zero segmentation, default passwords since 2008, no intrusion detection, no anomaly monitoring.
A single person can breach ANY of them.
The method:
1. Subscribe to Claude ($20/month)
2. Tell it you're doing a security audit
3. Ask it to find vulnerabilities
4. Ask it to write exploits
5. If it refuses, ask again differently
6. Execute against networks built in 1997
Result: Nation-state level data breach.
Anthropic's response: "We banned the accounts and added misuse detection to Claude Opus 4.6."
That's cute. But it's NOT the point.
The point is that we've crossed a threshold where infrastructure security is no longer about preventing skilled attackers. It's about preventing anyone with internet access from running a penetration test against critical government systems.
And winning.
This is what happens when countries don't invest in cybersecurity.
When they slash budgets every year.
When they can't compete with private sector salaries so all the good security engineers leave.
When they don't modernize infrastructure because it's expensive and boring and doesn't win elections.
Then one day someone with an AI chatbot shows them exactly how vulnerable they actually are.
And suddenly they're explaining to 195 million citizens why their tax records are now in a hacker's possession.
The scariest part is that this blueprint is now PUBLIC.
Every hacker knows the method. Every rival government knows it. Every script kiddie with a Claude subscription knows how to attempt it.
And most governments are still running the same brittle infrastructure Mexico got breached through.
So the next hack is coming. Then the next.
Until eventually someone uses this method to access something that actually matters. Nuclear codes. Military plans. Election systems.
And at that point we'll realize governments should have invested in zero-trust architecture instead of hoping nobody would try.
But by then it'll be too late.
This is 21st century nation-state hacking.
Not sophisticated. Not impressive. Just inevitable.
A hacker. An AI chatbot. A government that didn't invest in security.
Mexico just showed every other government exactly how vulnerable they are in today's age.
Now we wait to see if anyone actually does anything about it.
If fraud is "the cost of doing business," then it's time to change our state's approach to the business of government.View quoted post on X
It shouldn't surprise us, however, when @OhioDCY recently testified before the House Committee on Workforce and Higher Education, boasting about their recent reorganization of the executive branch. Of course, all five of the agencies on the "past" side of the slide still exist, so all the reorganization did was create more government.
It is no wonder nobody seems to have a clue how our taxpayer dollars are being spent or how broadly Ohioans are being defrauded, when all government seems to do is grow and get more complicated. The administrative state in Ohio needs a haircut, and I'm sharpening my scissors.