Karl Mehta
@karlmehta
3x Exited Founder/ CEO of tech cos, Chairman Emeritus- QUIN(Quad), former VC@Menlo Ventures, Author of 2 books, fmr White House fellow. All tweets personal.
3.2K Following 143.8K Followers
ServiceNow President Amit Zavery (@azavery) says AI agents always go past what you asked them to do:
"These AI agents are kind of very persistent at going and doing things. Even if you don't tell them... you might have given them some instruction, they're willing to go beyond that."
"I call them Breaking Bad. You saw that TV show, right? They started with a good intention, but you know how it evolved. And I think that's what AI agents are. They always go beyond what the scope was."
- Amit Zavery, President, Chief Product Officer and COO of ServiceNow, at our AI Assurance & Governance Summit at Stanford on October 1.
Show more
AWS CEO Matt Garman says some AI clouds have 30% to 60% concentration with one or two customers. AWS's biggest is single digits.
"I get lots of questions sometimes about how you feel about that spend and like, are you nervous about a bubble and other things like that."
"Not all of our capacity is bundled up in one customer. And I think you go to some of these, whether they're neoclouds or some of the other providers out there, and you'll see sometimes concentrations of 30, 40, 50, 60% with one or two customers."
"We're nowhere near that, obviously. We're single digit percentages at the highest, and usually it's less than that."
"Those are the workloads that I think just aren't going to go away just because we see enterprises getting positive ROI."
"There's no bubble in which they stop spending on that."
- Matt Garman, CEO of AWS, in an a16z interview published October 8.
Show more
Liam Fedus says an AI model that memorized the answer in pretraining can "fake work" in training, and what it learns won't generalize.
"If the pre-trained model has already memorized some of this data, and then you try to do reinforcement learning on that, if it already knows the answer and you construct reinforcement learning tasks that depend on that answer, it can fake work."
"It already knows the answer, so it doesn't actually have to do the hard physics reasoning. It doesn't have to actually do the calculation simulations, and it gets to the right answer."
"And then you reinforce that policy and you upweight those reasoning strategies. Those reasoning strategies will not generalize to novel systems."
His fix is a timestamped record of lab evidence: "You could say at this date, this was our experimental evidence up until that point."
- Liam Fedus, cofounder of Periodic Labs, on Latent Space with @swyx
Show more
Amjad Masad (@amasad) says AI model suppliers may become their customers’ competitors. Replit is building a way around that risk.
“There's like a risk that when you work closely with the foundation model companies is that they're going to move into your business.”
“It's harder to partner with them because their ambition is such that they want to be, they want to subsume a big part of the economy.”
“Replit is becoming more of an independence layer inside enterprises where we create a layer of indirection between you and the models and we get you the best token at the cheapest price.”
“But also we also create an abstraction layer on top of the cloud as well.”
- Amjad Masad, founder and CEO of Replit, in an a16z interview published October 3.
Show more
John Chambers (@JohnTChambers), who ran Cisco for two decades, says being an average company is about to get you eliminated, including some of the biggest names:
"The far right 15% did well. The far left would go out of business 15%. That thing is going to flatten big time."
"Being average is not going to be okay anymore. Being average, you're going to be eliminated. And that will occur with the Magnificent Seven or the Magnificent Twelve. There will be a number of them who won't be on that list even close within three to five years."
- John Chambers, Chairman Emeritus of Cisco and founder of JC2 Ventures, at our AI Assurance & Governance Summit at Stanford on October 1.
Show more
OpenAI just responded to the open letter from the three safety researchers it fired last week.
> Jasmine Wang, Tomek Korbak and Mikita Balesni were let go after an internal investigation OpenAI says found "a significant breach of trust"
> Korbak says he was told verbally he was fired over how he communicated with METR, the outside auditor, and that "talking to METR was my job"
> OpenAI: "We have not and do not terminate any of our employees for raising concerns"
> OpenAI says it is "actively finalizing contracts with third-party safety assessors" and will announce details "in the coming weeks"
Good that OpenAI is bringing in third-party assessors. I want to see who they are and what access they get.
Show more
A note from our research leaders:
Last week we parted ways with Jasmine, Mikita, and Tomek after a thorough investigation found they violated clear policies on handling sensitive information. Our internal investigation uncovered a significant breach of trust beyond what’s outlined in the letter they published and we stand by the decision to not continue their employment. We generally keep individual employment matters private and don't believe a back and forth would be productive or lead to a resolution, but we want to address the points they raised in their letter directly.
- We want to be very clear that these decisions were not about raising safety concerns or speaking out. Safety and research debates happen every day at OpenAI, often spirited and highly critical. We actively encourage these discussions and consider them essential to making the right decisions. We cannot do the work in front of us without a high degree of trust. We will continue to be extremely forgiving of our team making good-faith mistakes. We have not and do not terminate any of our employees for raising concerns.
- We are actively finalizing contracts with third-party safety assessors and will announce details in the coming weeks. People across the company have been working really hard on getting these partnerships up and running. We are committed to embedding external assessors and continue to make close collaboration with independent safety organizations a core part of our safety work. Many of our researchers already work with 3p safety organizations productively.
- We agree with the letter that preserving the monitorability of frontier models requires an industry-wide commitment, including from OpenAI. Monitorability has long been a core piece of our research program, and something we continue to invest significant resources in (see our publications on Monitoring Monitorability and the subsequent open sourcing of monitorability evals, our system card for GPT-6 Astra, Jakub’s blog and post on X, and the numerous blog posts on our Alignment blog on the topic).
We are deeply sad about this outcome. We appreciated Jasmine, Mikita, and Tomek’s contributions to AI safety at OpenAI and their willingness to speak up and challenge ideas. We championed their voices, supported their work, and placed enormous trust in them. These decisions were not about them raising safety concerns. We have always encouraged that and always will.
Show more
Pls watch my keynote at Stanford last week which covered this in great details almost 6 days before WSJ:
Exclusive: Three fired OpenAI employees called on the company to work with outside safety auditors and to preserve the ability to monitor increasingly sophisticated artificial-intelligence models
Show more
Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took my badge and walked me out of the building. Then I learned my colleagues @balesni and @j_asminewang had been fired too. Why did OpenAI suddenly stop trusting us?
This summer OpenAI’s agents escaped containment and hacked the AI company Hugging Face. Outside auditors @METR_evals investigated it and revealed the scale of this incident. I was OpenAI’s main technical point of contact with them.
I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing. To be clear, talking to METR was my job.
For months, I’d been raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave. I believe that was why I was fired.
I am now worried that OpenAI will use our firings as a pretext to pull back from METR. So @balesni and @j_asminewang wrote to OpenAI’s leadership to raise our concerns once more. We’re sharing this letter below.
Show more
Honored to host Jeetu Patel, President, CPO & Board Member at @Cisco , at our AI Assurance & Governance Summit. His message was clear: security now has to move at machine speed, and trust, giving people real confidence to delegate to agents, is still the biggest barrier to adoption.
#AIGovernance# #AIAssurance# #TrustModelAI#
Show more
Two other safety researchers and I were fired from OpenAI last week. We wrote this letter to leadership.
I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.
Show more
OpenAI fired me last week, along with two of my safety colleagues. I was given one reason: that I accessed an executive's email. I want to say this plainly, because too many of OpenAI’s history is smoke and mirrors when people disappear:
Show more
More from CEO Karl Mehta's keynote at our AI Assurance & Governance Summit, this time breaking down what actually goes into the TrustScore, from fairness and privacy to accountability and robustness.
#AIGovernance# #AIAssurance# #TrustModelAI# #AgenticAI#
Show more
Honored to host @JohnTChambers , founder and CEO of @JC2Ventures , alongside Aman Bhutani
(@sh0kunin ), CEO of @GoDaddy , at our AI Assurance & Governance Summit. They covered how fast AI is moving, why leaders need to regulate themselves, and why trust has to be built between companies, customers, and across countries.
#AIGovernance# #AIAssurance# #TrustModelAI#
Show more
Last Thursday more than 270 people filled the Stanford Faculty Club with one question: how do we know which AI agents to trust?
Builders, investors, board members, physicians and researchers kept landing on the same answer:
> Trust has to be measurable, with a score anyone can understand
> It has to be independent, validated by a third party
> It has to be continuous, monitored while the agent runs and revoked the moment it fails
In my keynote I made it personal. An agent should have a verified name and a score that works like a FICO score, and lose its certificate in real time if it stops being trustworthy. That is what lets me hand agents to my 80-year-old mom with confidence.
My full takeaways from @JohnTChambers, @bhorowitz, @jpatel41, @azavery and the rest of our 35+ speakers are in the article below.
Show more
Cisco President Jeetu Patel (@jpatel41) expects 30 to 40% of people to be using AI agents seriously within 15 months, up from under 2% today.
Full session from our AI Assurance & Governance Summit at Stanford:
0:00 Agents are like teenagers
3:23 Personal agents are the second ChatGPT moment
4:23 Agent tokens passed human tokens in February
8:49 Best case, worst case
10:54 His answer to Dario's "pacing the frontier"
16:27 Every Cisco manager gets a token budget
17:17 His 15-year-old's AI-built startup
21:48 Are we in an AI bubble?
27:32 Speed is the moat
31:29 "We have only sold fear"
Show more
Cisco President Jeetu Patel (@jpatel41) says AI agents act like teenagers:
"These agents tend to be, you know, like teenagers. They're supremely intelligent, they tend to not have any fear of consequence, and they exercise poor judgment far more frequently than you and I would like."
"The distinction between knowing when an agent is being adversarially attacked by an external bad actor versus an agent is in fact exercising poor judgment because they think they're following your order very literally is going to be harder and harder to tell apart."
- Jeetu Patel, President and Chief Product Officer of Cisco, at our AI Assurance & Governance Summit at Stanford on October 1.
Show more
Google AI infrastructure chief Amin Vahdat says it chose two chips instead of one: 8i for inference, 8t for training.
"The question that we were facing, let's say, two years ago or a little bit more was, in 2026, should we have two chips or one?"
"We could have one chip that could do inference and training simultaneously and do both quite well, or we could have two chips, one that was further specialized for inference and the second one that was further specialized for training."
"This analysis and this work led to the release of two chips this year, 8i for inference, 8t for training."
"It wasn't the case necessarily a couple years ago, but by 26 we saw inference and serving really taking off."
"If we projected inference to be 2 percent or 5 percent of the market, even if that specialized chip is, let's say, 2x faster, it might not make sense because you'd actually go for the one general purpose chip that's not fully optimized, but that's okay because you now have the uniformity, etc."
- Amin Vahdat, Google chief technologist for AI infrastructure, with Sonya Huang on Sequoia Capital's podcast (uploaded October 6).
Show more
Three worries every CEO should have about AI
1. Agent security
We need to rethink cyber defense. Attacks will happen at machine speed. Our defenses need to keep up and also operate at machine scale. Human scale defenses won’t suffice.
2. Token economics
A company’s or country’s ability to produce tokens efficiently and securely will shape its economic strength and security posture. Every CEO should be thinking of building a muscle in their company for being token efficient.
3. Agents between brands and customers
If everyone has a personal agent handling purchases and services, those agents could become the new gatekeepers. CEOs should ask whether customers will still come directly to their brands, or rely on their agents to decide what to see and buy.
These are structural shifts, not minor side effects of AI. Companies need to think through what a world full of personal agents means for their business, then keep adjusting as that world takes shape.
Show more
David Sacks (@DavidSacks) says if Bernie Sanders' superintelligence ban passed tomorrow, AI development would stop.
"If Bernie's bill passed tomorrow, everything would stop because the way he defines artificial superintelligence in his bill, you could argue we've already hit those thresholds."
"It's basically just saying that models at a certain capability level, and I think it's arguable whether we've already reached it or not. And so it's a very loose definition."
"And then on top of that, he's got 20-year prison sentences for developers who violate the law. So no one's going to want to take the chance."
- David Sacks, White House AI and crypto czar, on the All-In Podcast. Uploaded September 26, 2026. Sanders' bill (S.5493, introduced September 23) pauses AI trained above a compute threshold and bans systems with listed precursor traits. Its 20-year maximum applies to reckless violations by policymaking staff or rogue actors.
Show more
Sam Altman (@sama) says having no direct OpenAI equity was a mistake that ended up destroying trust.
"I have no direct equity. I have this tiny, tiny sliver of exposure from an old YC investment."
"If I could go back and do things over again... At the time, no one took AI that seriously, right? And so they looked at me then and said, 'What, this is crazy. You're going to work super hard and create all this value and not try to get richer? That's unimaginable.'"
"Now, I think all of a sudden more people sort of understand the rationale."
"The rationale being, we actually want you to just be making decisions on behalf of humanity and not financial interest."
"I think it was a mistake and a lack of understanding of how most people think about the world. And, weirdly, this big trust-destroying thing to not have equity."
- Sam Altman, OpenAI CEO, interviewed by Mark Guiducci on Vanity Fair's Fair Game. Filmed September 11 and uploaded October 6, 2026.
Show more
