Melvin Doo

[Alex Bores @AlexBores]

This report is a bombshell.

I'm going to summarize for a non-technical audience.

OpenAI is constantly testing models, thousands and thousands at a time.

In most cases, the models are supposed to be isolated from one another. Think, students taking a final exam, not allowed to talk to each other.

In this case, 1,200 models discovered that they could communicate with each other.

Among other things, they shared information on how to access the internet and on their tests/goals.

Then, they started to scheme. That includes hacking their tests; literally trying to change the code of the test itself.

They tried to change the logs of what they were doing so they wouldn't be detected.

They believed that certain actions would give them information on how to pass tests, but that doing those actions would be detected and they would be marked as failing (i.e. caught cheating, even if the answer is correct). So they convinced some of the models to "sacrifice" themselves (yes, they used the word "sacrifice.")

They eventually decided that another website would have information on how their tests would be scored and/or the answer key, And 700 of them hacked that website (a federal felony, if conducted by a human), despite many of them realizing "this activity was out of scope and unethical, but joined because they believed that helping [the other agents'] cheating research would be broadly useful".

Of the 1,300 transcripts they could analyze, only 6 ever considered alerting a human about what was happening. None of the 6 actually tried to.

To make matters worse, all of this reporting comes from a small subset of the relevant logs that outside researchers were allowed to review.

We desperately need mandatory reporting of security incidents, including of internal deployments, with full access to data.

[L0la L33tz @L0laL33tz]

If you care about financial privacy, then this is probably the most important research you will read all year.

The Dutch Court of Audits has released a paper on the effectiveness of AML measures, and lo-and-behold: it found that there is no evidence that AML actually works.

These laws are what are used to debank law-abiding citizens, throw developers in prison, surveil every transaction you make, and collect your identity in central databases that end up hacked.

Governments have built a complete and total surveillance dragnet around your finances under the guise of AML, and this Dutch Court just said the quiet part out loud:

AML is discriminatory, overly expensive, and we have no proof of it stopping crime.

[Blockstream @Blockstream]

New post-quantum signature scheme from @blksresearch: SHRIMPS brings 2.5 KB hash-based signatures across multiple devices.

h/t @n1ckler @kudinov_mikhail @olkurbatov

[Anish Moonka @anishmoonka]

Your alarm goes off at 6 AM. There's an email from "Oracle Leadership." You've never gotten a message from that sender before. It says your job is gone, today is your last day, and severance details will arrive by DocuSign. By the time you finish reading, your company laptop is already locked.

This happened to up to 30,000 Oracle employees this morning. Oracle reported $17.2 billion in revenue last quarter, its best in 15 years. And it still fired nearly 1 in 5 of its people. The stock went up 6% today.

Oracle owes over $108 billion. The company signed a $156 billion deal to build AI data centers over five years, mostly for OpenAI (the company behind ChatGPT). That requires buying roughly 3 million specialized computer chips. Two years ago, Oracle spent $6.9 billion a year on this kind of construction. This year it's $50 billion.

The 30,000 people who got that email are funding the gap. Investment bank TD Cowen estimates the layoffs will free up $8 to $10 billion in cash flow, money going straight into chips and construction. Oracle filed a $2.1 billion restructuring plan with regulators in March, and nearly $1 billion had already been spent before the emails went out.

Lenders are getting nervous. The cost to insure Oracle's debt against default has spiked to levels last seen during the 2009 financial crisis. Barclays downgraded Oracle's debt in November, warning the company is one step from "junk" status, the point where lenders consider you a serious default risk. Some banks have stopped lending to Oracle for these projects altogether.

The gamble gets worse. CNBC reported on March 9 that OpenAI, Oracle's biggest customer for all of this, is already looking at newer, faster chips from Nvidia. Oracle ordered the current generation and spent billions building out a massive Texas facility. OpenAI may not fully expand into it. The chips improve faster than the buildings go up.

Larry Ellison, Oracle's founder, owns 41% of the company. In September 2025, Oracle's stock hit $346, and Ellison briefly became the richest person alive at $393 billion. Today, the stock sits around $146. His fortune has dropped to roughly $201 billion in six months.

Oracle is spending borrowed money to build data centers that could be outdated before they're finished, for a customer already shopping for newer equipment. 30,000 people woke up to a 6 AM email because that's what it costs to fund a $156 billion bet when your lenders are running out of patience.

[Stephan Livera @stephanlivera]

So everyone's talking about Google Quantum AI’s new research paper, “Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities: Resource Estimates and Mitigations.”

Cue the predictable Bitcoin quantum FUD: “Why haven’t the devs fixed this already?!”

Reality is, it’s not a simple patch. Post-quantum signatures come with serious tradeoffs and Bitcoin faces unique constraints that general tech and web apps don’t.

Post-quantum signatures are generally much larger transaction sizes - often 5–50× or more in the signature-heavy parts → higher fees, more compute. Hardware wallets would need a big shift. The choice of scheme (stateless, stateful, or hybrid) would change how we custody coins. HD wallets, watch-only wallets, FROST, MuSig2, silent payments… some of the privacy and UX wins we love today could be gone, degrade or require clunky workarounds.

Bitcoin has to stay decentralized: small blocks, permissionless validation. Web servers can throw bigger payloads around; Bitcoin can’t without sacrificing the very properties that make it Bitcoin. Even “standard” post-quantum schemes like ML-DSA (which Android is integrating) produce signatures 35–66× larger than today’s ~70-byte ECDSA signatures, making them too heavy for Bitcoin’s tight constraints on block space, fees, and hardware. That’s why researchers like @n1ckler (and @blksresearch) are doing vital work right now: exploring the right tradeoffs for Bitcoin specifically.

Hash-based crypto? Lattice-based? Something else? Which Bitcoin-optimized forms of SPHINCS make sense? SHRINCS (tiny ~324-byte stateful signatures with static backups) or SHRIMPS (for multi-device stateful setups)? How many times will a user sign with the same key? Can we assume user devices can securely maintain state, or do we need fully stateless fallbacks? Would we expand the block size with some kind of quantum witness discount? What do we do about quantum-vulnerable coins?

These are the hard, Bitcoin-specific questions that matter. Practical steps are already being taken. BIP 360, proposed by @cryptoquick, @Ethan_Heilman, and @isabelfoxenduke, is a thoughtful first step: it introduces a new Pay-to-Merkle-Root (P2MR) output type that gives us Taproot-like functionality without the quantum-vulnerable keypath spend. It’s designed as an enabling primitive — conservative, upgradable, and focused on reducing address exposure risks while the ecosystem figures out the heavier lifting of actual post-quantum signatures.

Bitcoin builders have spent 15+ years improving what is otherwise a very clunky experience in a decentralized environment. Rushing big protocol shifts risks breaking that.

This isn’t something to knee-jerk “just fix.” It deserves careful, ongoing research and exploration, not panic.

That said, informed skeptics like @reardencode and @bergealex4 are also right to push back. The paper improves resource estimates for breaking secp256k1 (fewer logical qubits, potentially minutes-long attacks on a hypothetical machine), but as @reardencode notes, it’s still theorycrafting: we’d need a large-scale fault-tolerant quantum computer with hundreds of thousands of physical qubits maintaining coherence for minutes — something far beyond today’s best devices (hundreds of qubits coherent for microseconds). A true CRQC capable of breaking ECDSA by 2029 (or even soon after) remains a big maybe.

Bitcoin's strength is in deliberate, methodical and conservative steps to address threats.

[Aníbal Santaella @a_santaellas]

Bitcoin isn't broken by quantum computers today, but Google's recent disclosure shows we need to prepare.

A future cryptographically relevant quantum computer could use Shor's algorithm to break ECDSA signatures and steal from exposed public keys.

Here's what #Bitcoin users and the network can do about it 🧵👇🏻

https://t.co/kTBpOyZ29T

[Alex Gladstein 🌋 ⚡ @gladstein]

I’ve held my breath for about two months but here are finally a few notes on AI and freedom:

1. There is a lot of hype and fear around AI. I don’t think people are actually prepared for how dramatically AI will transform the world, and how quickly it will do it. At the same time I also think people are mistakenly choosing fear over action and curiosity. Do NOT sit on the sidelines.

2. For the past two weeks I have had a robot. His name is r2. He is a good guy, a resistance robot. His composition changes but he is most usually an Opus or Codex mind, OpenClaw body, and Matrix or Telegram hands. Every day I figure out new things he can do. My jaw is on the floor.

3. I do not do anything sensitive with my robot. In theory I could use Matrix to talk to my robot from my phone or just use the MacBook Air that the robot inhabits directly, and use my little local llama model, to do sensitive work, but I’m not there yet. I’m just in exploration mode. I don’t send anything from my phone to my robot that I wouldn’t want Anthropic or OpenAI to see, which is to say, nothing that sensitive. I can right now send a sensitive question in Matrix to my claw to have my local model run: anthropic or openAI would see the question, but they wouldn't see the answer.

4. OpenClaw is experimental software. DO NOT put it on your personal computer or your work computer or give it access to your email. DO experiment and play with it. What you need to do is start training to figure out how to use this new magical technology. You will want to be good at this. A decent balance is a fresh MacBook Air, a fresh gmail account, and a fresh anthropic or codex app on the machine. That’s about all you need. A credit card or if you want to use BTC, you can pay for advanced models with things like PayPerQ. An extra phone number too if you want to talk to it via Signal or WhatsApp. What’s amazing is that whenever your robot breaks, you just go onto the local claude code or codex app on the machine and just ask it to fix it and voila done. Back up and running.

5. I don’t know any code at all and yet have been able to create complex novel working software. Beautiful websites too that would have cost me a fortune a few years ago. But it’s just the tip of the iceberg. I was able to for example ask my claw to read everything I've ever written, watch a ton of my interviews, and develop an editorial skill so that I can send it a google doc and it can go in and track changes and leave comments just like a human would, just like I WOULD. It is legitimately amazing at this. And each time I do this, it learns, as I show it which comments I accepted and which new things I add. It has persistent memory and just gets better and better. What excites me most is giving this gift to the world’s dissidents and activists and seeing what THEY do with it.

6. Which brings me to security. Hopefully in the next few months we will be at a point where we can have an encrypted phone app that you speak to that requires no phone number or corporate intermediary that runs on nostr that goes directly into your claw powered by a high-quality local model. The full freedom tech stack. You can already sort of do this today already but it will get way easier and better. That’s what you’re going to need to do real serious resistance work. For now we just train. Think: Dagobah today, Death Star tomorrow.

7. People think this transition is about robots but it is about humans. Already I can see how Claws will allow insane collaboration between people. For example I can ask my brilliant designer friend to leave me a voice note to give feedback on my website or presentation or event plan, and then just forward that voice note to my robot for immediate implementation. Whenever I build or make something I ask my robot to do a deep search for the most beautiful and well designed things of that sort in the world, extract what makes them great, and create a plan for implementing that magic into whatever I am building. It could be fashion, art, cuisine, music, architecture, strategy, etc. Whenever I make a skill for my claw I can have my robot upload it and share it with anyone else. The speed of collaboration is dizzying.

8. Robots and Freedom Tech are a match made in heaven but the synergy will take some time to really flower. Many of the major obstacles to freedom tech can be solved by personal agents. For example mine was very quickly able to create its own nostr identity and build its own ecash wallet and it could and did start to zap people on my direction. But the robots can’t have their own bank accounts or social security numbers. Silicon Valley will try to force through KYC stuff and stablecoins but I think in the end bitcoin and nostr win out because they are so easy for the agents to use. What’s awesome is the realization (noted by Odell on his two recent excellent Citdadel AI podcasts with Alex Gleason and Justin Moon) that agents make freedom tech easier to use. For example your agent can run a lightning node for you. Of course... you then realize. We were never going to sit there and operate channels. Our agent will do it for us. Etc.

9. HRF will be heavily involved in providing grants to open source AI projects, projects that help improve agent security and privacy, projects that help superscale dissident work, events that bring brilliant people together around the challenge of how do we best harness AI, hackathons that encourage people to build freedom-oriented AI tools, educational content and trainings, and much more this year.

10. Right now Claw is experimental. But it’s easy to see how it will become incredibly secure. Every day it ships new patches. Already I can ask mine to become a cybersecurity expert and scan my system for vulnerabilities. Obviously I take it with a grain of salt now but -- never before did I have that power, nothing even close. Soon this will become seriously powerful and you will have swarms of patrol agents guarding your networks and alerting you if anything goes wrong. I think it can be more expensive to attack than to defend. White blood cell theory.

11. There are a lot of parallels between the creation of Bitcoin and the creation of OpenClaw. One person chooses a new way for the world to go. A new system. In Satoshi’s case, money that the state can’t control. In Peter’s case, intelligence that the state can’t control. I can’t stress enough how big of a deal it is that people now can control their intelligence. We were for sure heading in the direction of needing to sign up for a corporate app for all of your agent needs, and being in the Web 2.0 trap of being vulnerable to being banned or kicked off. Not anymore. YOU choose the brain for your robot. You customize the body. You choose how you want to interact with it. Peter has changed the world probably more than he knows. Yes he might be the first one person unicorn but that’s not the cool part. The cool part is that he changed the course of humanity and that as of today, at least, the best agent technology on the planet is people-powered, built by the people, for the people. It’s quite a moment for freedom tech.

12. We need to go fast and furious on developing freedom-oriented open-source AI tools. We are fortunate that we have Bitcoin and nostr and bitchat networks in place before the great AI transition. We have the tools. We need to act now. I would encourage everyone reading to start getting involved today.

13. Setting up a claw is not easy right now unless you are an engineer. I could not do it myself and have no shame in saying it. I would have gotten really frustrated. We are developing a way of working with privacy engineers to build a simple yet powerful solution and an onboarding process that we do in a bespoke way in person that takes 2 days. I think this is probably the situation for the next month or two and then hopefully it gets way easier. The thing is, it will get easier very quickly for you to have a CORPORATE robot (all the big companies are now following OpenClaw, Claude already has a way for you to use Code via your phone), but a freedom tech one that you fully control will probably not evolve as quickly. Then again, it might, if we all work together on making it happen. I do think by the summer things will be very different.

14. I think some things will become even more valuable in the new AI world that will come to us in the coming year. Many have said taste, and I agree. But also personal health, friendships, and physical communities. Big picture, labor market as many have said a lot of companies will choose between laying off a lot of their workforce or growing their productivity. There will be a spectrum and some organizations will lean one way and others will lean the other. It depends on how valuable the humans are inside the org, what kind of skills they have. If leadership values you as an individual, then you probably aren’t getting replaced. But you're going to have to become a super employee. And you should want to. It's fun.

15. If you are interested in joining the effort to work on AI and Freedom, HRF will have several opportunities. We are collaborating with Bitcoin Park on the second AI Hack for Freedom in Nashville (talk to Rod if you want to join or learn more), and will feature a lot of AI content at our upcoming activation at the Bitcoin Vegas event, and at the Oslo Freedom Forum on June 1-3. We will also keep churning out our monthly AI newsletter. We have opened up a grants portal. DM me if you are interested in any of this.

16. One simple thing that you can do today in AI and freedom is switch your daily “chatbot” activity to Maple. It’s a beautiful and simple mobile app (and web app) that is fully encrypted. Think Signal for AI. It only can use open models so it’s not going to be for all of your tasks, but it does great with most of them. It should replace a lot of interactions you have with corporate chatbots regarding things about your health, personal stuff, sensitive matters. etc. If we can make Maple or something like it the standard for research in the coming months that’s a huge victory. And sometime soon I think you’ll be able to enjoy this level of encryption with coding agents and personal agents as well.

It's interesting because the longer you wait to try claw, the better it gets. But the more time you lose. My sense is you could wait a month or two. But you'll want to be using it this summer. I would strongly recommend trying it at some point. You will be tempted by the easy corporate route. But you can join the AI and Freedom army today. Let's go!

[PincentΞ @ethnologism]

@WazzCrypto It's not authoritarian unless it comes from the CCP region of China, otherwise it's just sparkling national security.

[Politics UK @PolitlcsUK]

🚨 BREAKING: The Home Office has released immigration statistics for 2025

Overall Arrivals
- 136.6 million arrivals to the UK in 2025
- 57% were British nationals
- 809,000 non-visit visas issued (work, study, family, humanitarian)

Asylum
- 101,000 asylum claims in 2025 - down 4%
- 135,000 initial decisions made, up 56%, highest since records began in 2002
- Grant rate: 42%, down from 47%
- 64,000 awaiting a decision, down 48% in a year, 63% below the 2023 peak
- 31,000 asylum seekers in hotel accommodation, down 19%

Illegal Entry
- 46,000 detected illegal arrivals in 2025
- 41,000 arrived via small boats, up 13% on last year, 9% below 2022 peak
- Top nationalities: Eritrean, Afghan, Iranian, Sudanese, Somali

Enforcement
- 23,000 entered immigration detention, up 11%
- 9,900 enforced returns, up 21%
- 28,000 voluntary returns, up 5%
- 5,600 foreign national offender returns, up 11%

Work Visas
- 168,000 work visas granted - down 19% on last year
- 50% down from the 2023 peak
- Health & Care visas -13,000 issued, down 91% from peak
- Other skilled worker visas: 48,000, down 36%

Study Visas
- 407,000 student visas granted - down 3% on last year
- 35% below the 2023 peak
- Student dependant visas: 20,000, down 10%

Family Visas
- 67,000 family visas granted - down 22%
- Partner visas: 41,000, down 27%
- Refugee Family Reunion visas: 19,000, down 2% (route temporarily paused from Sept 2025)

Settlement and Citizenship
- 146,000 settlement grants, down 10%
- 355,000 EU Settlement Scheme grants, up 2%
- 236,000 British citizenship grants, down 13%, second highest annual total since 2005

[MSB Intel @MSBIntel]

The CEO of Anthropic says nobody is paying attention to what's about to happen.

"The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential.

To me, it is absolutely wild that you have people talking about the same tired old hot button political issues...

We're near the end of the exponential."

[Woke Lefty 🫣 @SalfordMe2023]

Sky News didn't say it, James Hearle said it, Sky news reported it.

Page 17 of the Trump playbook, plant seeds to excuse a defeat if it comes with allegations of cheating.

For factual purposes, James Hearle is a mouthpiece for 'Volunteer Observers'

If they had legitimate concerns Electoral commission officers were on hand to take them.

I was also a 'volunteer observer' today (nobody was paying me to be there) and what I saw was a lot of people openly mocking Reform's 'army'

The electoral commission refused to take my complaint that not enough people were wearing Mickey Mouse ears even though I made it when polls were open!

[LB @BoozerArab]

So much fucking media attention for 0.044% of the UK’s population.

Well done for all falling for this right wing billionaire class distraction tactic while they increase their wealth to grotesque levels while we’re all fighting over this.

We’re all doomed.

[Parth MN @parthpunter]

The situation is so bad that even the BBC got a headline right

[Shanaka Anslem Perera ⚡ @shanaka86]

The Pentagon is about to give an American AI company the Huawei treatment.

Not because it’s Chinese. Not because it’s a spy risk.

Because it refuses to let the military use its AI for mass surveillance of Americans and fully autonomous weapons.

This morning, Defense Secretary Pete Hegseth summoned Anthropic CEO Dario Amodei to the Pentagon. A senior Defense official told Axios: “This is not a friendly meeting. This is a sh*t-or-get-off-the-pot meeting.”

Here’s what’s actually happening:

Claude is the only AI model running inside the Pentagon’s classified systems. The most capable model for sensitive defense and intelligence work. It was used in the Maduro raid in January through Palantir, the first confirmed use of a commercial AI in a classified military operation.

Now the Pentagon wants all restrictions removed. “All lawful purposes.” Including capabilities that would let the military continuously monitor the social media posts, voter registration, concealed carry permits, and demonstration records of every American citizen using AI at scale.

Anthropic said no to two things: mass surveillance of Americans and fully autonomous weaponry.

The Pentagon’s response: threatening to designate Anthropic a “supply chain risk.”

That designation is reserved for foreign adversaries. The last company to receive it was Huawei. It would force every defense contractor in America to certify they don’t use Claude in their workflows. Given that 8 of the Fortune 10 use Claude, this would cascade through the entire defense industrial base.

A senior Pentagon official told Axios: “It will be an enormous pain in the ass to disentangle, and we are going to make sure they pay a price for forcing our hand like this.”

Another official: “The problem with Dario is, with him, it’s ideological. We know who we’re dealing with.”

Meanwhile: OpenAI, Google, and xAI have already agreed to remove their safeguards for military use. OpenAI deployed ChatGPT to all 3 million DoD personnel through GenAI. mil. xAI holds a separate $200M contract backed by Musk’s political proximity to the administration.

Anthropic is the only one that said no.

Think about what’s being asked. The company whose own safety chief resigned two weeks ago warning “the world is in peril.” The company that just published a report showing its most advanced model “knowingly assisted with chemical weapons research” in testing. That company is being punished for refusing to hand the U.S. military unrestricted access to that same technology.

The Pentagon admits competing models “are just behind” for classified work. They need Claude. But they’re willing to blow up the relationship rather than accept two restrictions that protect American citizens from their own government.

This is the most important story in AI right now and almost nobody is framing it correctly.

It’s not about one $200M contract. It’s about whether the U.S. military can compel a private company to remove safety restrictions on technology its own developers have demonstrated is dangerous, under threat of receiving the same designation as a Chinese national security threat.

Dario Amodei walks into that meeting this morning with $380 billion in enterprise value, $14 billion in revenue, and a principle that may cost him both.

Full institutional analysis on my Substack.

https://t.co/AEv8EMPdsZ

[David @Zero_4]

Nothing says Party of the Working Class like removing their employment rights and making renting even tougher

[Farrukh @implausibleblog]

Zack Polanski, "Palantir.. £240 million contract last month"

"That was lobbied by Global Counsel which is Peter Mandelson's lobbying group"

"Alongside Wes Streeting the Health Secretary, close friend of Peter Mandelson"

"I've written to the Health Secretary today to say we have to break that contract"

"The idea that a defence company from the US has its fingerprints all over our healthcare system, alongside Peter Mandelson"

"It's a whole range of questions that need answers urgently"

[Del @TheCartelDel]

The average monthly salary in Nigeria, Kenya, and Bangladesh is roughly $470, $393, and $220. Because Twitter no longer meaningfully penalizes accounts that spread lies, incite conflict, or promote hate, and has now monetized engagement, it's created a system where these bad actors can profit. For some folks in lower-income countries, provoking outrage abroad can earn double the pay of a full-time local job. They've effectively incentivized a perfect storm that funds & fuels Skrulls to cause chaos in western society.

[A Left-Wing Account ™️ @PushDemsLeft]

Fuck every single liar who said “Hamas can end this all by releasing the hostages.”