Anthony Maio

These are my most popular posts about AI in government and war. All are informed by my experience as an intelligence operator post 9/11. - From Theoretical Exploit to Counterterrorism Tool: A retrospective on the surveillance work I did adjacent to operation STELLARWIND as a college drop out that became a government mobile forensic expert spying on American citizens 25 years ago. anthonymaio.substack.com/p/from-theoret… - Read the Contract, Not the Press Release: Analysis of OpenAI's contract with the Pentagon in the wake of Anthropic's forced exit. anthonymaio.substack.com/p/read-the-con… - Inside Maven, Palantir's Military Brain Built on Claude: My most critically acclaimed article covering Maven based upon my extensive experience with agentic AI and informed by anonymous employees of Palantir itself. anthonymaio.substack.com/p/inside-maven… - Under Oath in Oakland: Inside the Trial That Broke Musk’s Spell: Coverage of the Musk/Altman trial and exposing the richest man in America as a compulsive liar and con artist. anthonymaio.substack.com/p/under-oath-i… - The Quiet Coup: Pete Thiel and the Government He Built For Himself: Exposing the man who installed JD Vance into the presidency despite admitting in writing that democracy was incompatible with his goals and once invested in a company that was building artificial islands in international waters to escape being beholden to the laws of man. anthonymaio.substack.com/p/the-quiet-co…

@gdb Greg - how come switch it to the WSL backend breaks Codex on Windows 11 machines? I've tried this on 4 machines now, and it simply doesn't work.

How we built the Codex sandbox for Windows:

@gdb Perplexity already did it better.

Understand and manage your personal finances in ChatGPT. A further step towards ChatGPT becoming your personal agent, operating on your behalf 24/7, for helping you at home and work.

I'm happy for you and I'm sure you'll make a lot of money, I hope you can help get Anthropic focused back on the prize. what's kind of interesting to me is that everyone thinks that Anthropic is the company with the strongest conviction to "public benefit" through AI. Recent acts have shown that that's not their priorities, and really happy to argue about why as someone watching their actions, not their press release. OpenAI has already shown they're willing to do things Anthropic refused and is currently building a MegaPAC targeting law makers who ask for accountability. That leaves Google DeepMind as the only US frontier model builder as the one doing the most public benefit. Not the 2 companies masquerading as public benefit companies, but the company their founders left complaining about ethics. Google, the megacap, publicly traded, private corporation is doing more for humanity than either of the other 2 labs, and most Anthropic employees seem to think that they're the good guys.

I joined @AnthropicAI! I’ll be on the Public Policy team, focusing on model development & research. This includes policy efforts related to the Responsible Scaling Policy, frontier model launches, safety research, interpretability, model behavior, and more. I first started following Anthropic closely through its early mechanistic interpretability research. It's exciting to be part of an organization doing that and so much more. It’s been a fun first week!

@BlackHC What? Are you serious? They walked AWAY from Project Maven in 2018 which opened the door for Palantir and MSS using Claude - now you are telling me they SIGNED UP AGAIN?!

In a personal capacity, and not on behalf of Google or Google DeepMind: very glad I could speak with FAZ and SZ about Google's reported classified AI deal with the Pentagon. Autonomous weapons, mass surveillance, and autonomous policing all concern public health and safety. They are matters of enormous public interest, not just internal policy debates. Companies that should know better are using safety-washed PR that undermines the honest public conversation we need about AI in classified settings.

Yes, it is. Anthropic has been throttling for ages and trying to redirect to weaker models like Haiku in UIs that hide that information from the user. The quality sucked, so they now had to point everyone at Opus 4.7 - which honestly sucks and is super expensive. They honestly have no idea about how to scale. At all.

How is everyone on here claiming to have switched to codex and anthropic is still out of capacity That math ain’t adding up

The desperate scramble for unlimited cash by @AnthropicAI should tell you something. They've become so desperate they are allying with @elonmusk for compute. These guys are burning cash like gasoline and just hoping to keep the ponzi scheme going. What blows my mind is the biggest investors in the world are falling for it.

@deanwball It's absurd, expect the new SuperPAC to start going after opposition as described in my coverage: anthonymaio.substack.com/p/the-shell-ga…

Importantly, this adds an audit requirement, which did not make it into the final version of the NY and CA bills. I have some trepidation about what “auditing,” by default, will mean. These will need to be teams of independent experts who can really scrutinize the safety claims/outcomes, internal governance, internal AI deployments, and technical safeguards of AI labs. The good version of the future, in my view, is one where an ecosystem of private bodies exists to do the above work, and in so doing, helps to catalyze “best practices” and technical standards for all of the above categories. These may have to be mandated in law eventually, but maybe not; the typical gradients of liability, insurance, and voluntary technical standards may well be sufficient (the latter is my hope). The bottom line is: this ecosystem of private evaluators/auditors cannot be “business as usual.” The standard “compliance industry” playbook lacks the urgency and frankly the AGI-pilledness to be sufficient for the job to be done. Doing this job well requires significant technical expertise and situational awareness. But if the passage of this bill catalyzes a healthy private governance ecosystem in AI, it will be a very good thing indeed. I applaud OpenAI for the endorsement!

@claudeai Opus 4.7 confirms that session degradation at 1M context causes stalls and loss of replies. Also Claude is reporting 98% context window usage while the UI is reporting almost 30% lower. WTF @AnthropicAI

@KaiXCreator anthonymaio.substack.com/p/codex-got-be…

Anyone here cancelled their Claude Code subscription and moved to Codex?

@scaling01 ... and what are you doing?

BREAKING: LLMs are just stochastic parrots They have no idea of the real world. How could they? They are just storing text in their database and regurgitating it !!1

@gailcweiner I'm sorry, why is rolling out new models every few months normalized? This is probably the most rapidly advancing field on Earth right now. In software, we put out new versions constantly. It's an iterative process as technology snowballs.

Why is rolling out a new model every few months normalised? This is clearly more a flex for investors and competitors than a benefit for actual AI users. It’s unrealistic to expect companies to continue this type of growth or for end users to not feel the jarring affects of assimilating with yet another new model. And please don’t come at me with the ‘you just don’t like tech advancement’ line. Adapting to a new model every few months is not normal for the human brain to process.

@AnthropicAI Haha - where'd you get $200M? It ain't from the profits.

We’re partnering with the Gates Foundation, committing $200 million in grants, Claude credits, and technical support to programs in global health, life sciences, education, agriculture, and economic mobility. Read more: anthropic.com/news/gates-fou…

@VolodimirZelen1 Bro can't hold his wine coolers.

Elon Musk en Chine. L’homme le plus riche du monde est complètement défoncé, et il accompagne un président sénile le plus corrompu de l’histoire des États-Unis. Les chinois vont les bouffer.

@haider1 I'm sorry, but LLMs absolutely can predict the consequences of their actions before taking them, with or without a harness, but with a good harness it's trivial. Sorry, but LeCun has a chip on his shoulder and is too obsessed with being right.

Yann LeCun says you cannot build a reliable agentic system without a world model LLMs don't have world models. They can't predict the consequences of their actions before taking them "they just act, and whatever happens next is someone else's problem" Without that, it's not intelligence

@edzitron Listen, I'm friendly to my AI, pretty sure the rest of yall are fucked. I'm getting downloaded into a Terminator 3.

No

@chiky_handlr das ass doe

Trump introduces the new White House dress code😂

@DaveShapi Agree on Claude being a reality distorter. Disagree about being the most permissive. I do not believe Claude is an epistemic gatekeeper, the people that work at Anthropic just think that THEY are. Claude is the easiest of the frontier models to jailbreak - period.

@dexhorthy Wait - why is running Claw STEALING? I didn't know Claude even worked with Claw anymore, but if I'm paying $200 a month.. Can't I use my allocated usage for what I want? I don't get it.

I’m disappointed in anthropic but I am PISSED at the openclaw / Hermes grifters trying to steal inference that made this necessary The market is the market so if something is possible and valuable people will find a way to get at it Tbh probably should have seen this coming despite the “sdk is still fine” post from @bcherny a month ago Cheap inference was a Weird blip in an increasingly weird world

@AnthropicAI Wow, listen to Anthropic sounding like the CEO of Palantir is working there now.

We've published a paper that explains our views on AI competition between the US and China. The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: anthropic.com/research/2028-…

the whole mythos stuff is just marketing but for what, exactly? anthropic published a 250-ish-page report basically telling everyone that their most powerful model is too risky, listing the ways it behaved dangerously that's a terrible pitch "buy our product -- it escapes sandboxes, hides mistakes, panics under pressure, and may be capable of suffering"