Price:

AI & TECH

US security intervention poisons Anthropic’s safety play

Saturday, June 20, 2026 · from 4 podcasts, 8 episodes
All-InRabbit Hole RecapAI Daily BriefBreaking Points
  • The US government shut down Anthropic’s models after the company’s warnings about AI danger became its own pretext.
  • Export controls ban foreign nationals, including key researchers, creating a citizenship barrier to intelligence.
  • The unilateral move destroys US reliability as a tech partner and accelerates global sovereign AI efforts.

Anthropic CEO Dario Amodei spent months arguing frontier AI models were too dangerous to release without government oversight. The Department of Commerce took him at his word. Citing a reported jailbreak discovered by researchers at Amazon, the agency issued an export control directive at 5:21 p.m. Eastern on Friday, forcing Anthropic to suspend all access to its flagship Fable 5 and Mythos 5 models for any foreign national worldwide.

The irony is total. On The AI Daily Brief, Nathaniel Whittemore notes that by framing its technology as a potential weapon, Anthropic gave the state the moral high ground to seize control of its product. AI entrepreneur John Enis argued the company’s safety rhetoric was largely a strategy for regulatory capture and masking compute scarcity. The lab’s cynical plan backfired.

"They screwed us."

- Administration official, The AI Daily Brief

The directive’s immediate impact is a new caste system for intelligence. Key Anthropic technical staff, including high-profile researcher Andrej Karpathy, are non-U.S. citizens on visas and are now legally barred from accessing the models they helped build. On Breaking Points, Krystal Ball argued the move reflected political favoritism, with Anthropic singled out as a disfavored lab that refused to 'bend the knee' while rivals like OpenAI and xAI operate with relative freedom.

The government’s legal theory is aggressive. Commerce Secretary Howard Lutnick reportedly phoned into a meeting with Anthropic’s technical team from the G7 summit, telling the company that if the model couldn’t stay online under the foreign national restrictions, 'That's the point.' Charlie Bullock of the Institute for Law and AI said the government's position is vulnerable to legal challenge, but Anthropic will likely settle without litigation.

The technical justification appears weak. Anthropic contends the jailbreak method finds vulnerabilities already widely discoverable via other public models like OpenAI’s GPT-5.5. Leda Security CEO Katie Moussouris argued the behavior - refusing to review insecure code but generating patches - is standard cyber defense. Over 100 cybersecurity experts signed an open letter stating removing Mythos from defense tools actually increases vulnerability.

"This is a major turning point for AI regulation."

- Aaron Levy, The AI Daily Brief

The precedent chills the entire industry. If the standard for a shutdown is a 'non-universal jailbreak,' no lab can safely ship a frontier model. On All-In, David Sacks claimed Anthropic believes AI is dangerous and competition is a 'race condition,' aiming to centralize control into a government-anointed cartel. The government now views these models as strategic assets, not consumer software.

The strategic cost outweighs any security gain. By pulling the plug on a commercial model without warning, Washington signaled to every ally that relying on American AI is a national security risk. Analyst Gail Weiner argued procurement officers in Tokyo or Brussels now have an unassailable argument to move toward sovereign AI or open-weight alternatives like DeepSeek. The U.S. narrative of being a predictable, rule-of-law provider evaporated overnight.

Analysts warn the move may pop the AI bubble. If the government can cap the monetization of the most powerful models, the massive capital expenditure required for the intelligence explosion loses its logic. Deutsche Bank’s Jim Reid warned a prolonged ban threatens U.S. tech firms and AI adoption speed, as you cannot rely on technology that could be switched off. The honeymoon between AI labs and the federal government is over.

#Regulation#Big Tech#Models
Andrej KarpathyDario AmodeiHoward LutnickNathaniel WhittemoreAmazonAnthropicOpenAIxAIGPTGPT-5

Source Intelligence

- Deep dive into what was said in the episodes

All-In with Chamath, Jason, Sacks & Friedberg
All-In with Chamath, Jason, Sacks & Friedberg

World's First Trillionaire, Anthropic Fable Banned, The New Oligarchs, Iran Peace Deal • Jun 19

  • David Sacks argues that Elizabeth Warren, Bernie Sanders, and Ro Khanna are forming a new oligarchy to dictate the economy, capital allocation, and labor allocation, masking their control under claims of equity and justice.
  • David Sacks claims that government-provided benefits create learned helplessness and reduce economic mobility, pointing to his family's experience on welfare as evidence.
  • David Sacks claims the Pritskar law in Illinois signals the erosion of private property rights by allowing government to tax assets annually beyond income taxes.
  • David Sacks clarifies that Elon Musk's wealth increase from the SpaceX IPO is paper wealth tied to stock valuation; he has no additional cash, remains under a one-year lockup, and plans to hold shares longer.
  • David Sacks argues wealth creation stems from building corporations as 'machines that make stuff', not from accumulating assets, and that tech companies enable workers to transition from labor to capital ownership.
  • David Sacks recounts the Fable 5 shutdown: Anthropic expanded Mythos preview to unauthorized parties, a partner found a jailbreak, and Dario Amodei refused a government request to take it down until receiving an export control letter.
  • David Sacks claims Anthropic believes AI is dangerous and competition is a 'race condition', aiming to centralize control into a government-anointed cartel, which captured Biden's AI policy officials.
  • David Friedberg argues the AI market will naturally fragment across hardware, clouds, models, and applications, driven by competition, similar to the disaggregation of IBM's mainframe monopoly.
Also from this episode: (1)

Diplomacy (1)

  • David Sacks states the proposed Iran deal reopening the Strait of Hormuz, securing a ceasefire and Iran's nuclear material removal, and costing the US nothing represents a far better alternative than sending ground troops.
#Models#Big Tech#Space#Markets#Macro
Rabbit Hole Recap
Rabbit Hole Recap

RABBIT HOLE RECAP #414: BITCOIN IS THE BEST MONEY • Jun 18

  • Block’s new AI tool, BuilderBot, now merges about 50% of the company's production code changes, handling 1,500 pull requests per week by researching, writing, and testing code autonomously.
  • Robinhood is cutting 10% of its full-time workforce, about 290 roles, to flatten management and operate more efficiently as its stock lags the broader market.
  • The hosts claim China's Gege Networks is developing AI tools to predict political dissent by building behavioral profiles from social media, location, and telecom data.
  • Marty argues the forced removal of Anthropic's Fable 5 model is a mix of political retaliation and a potential pretext for establishing a KYC/AML licensing regime for AI models.
  • The Trump administration is backing XAI against an NAACP lawsuit over data center emissions, framing the AI infrastructure buildout as a national security priority.
  • SpaceX is acquiring AI code tool Cursor at a $60 billion valuation, a move Bill Aamann argues is strategically accretive due to SpaceX's high market value attracting talent and enabling cheap acquisitions.
  • A leaked list reveals over 200 global elites, including tech founders and politicians, are members of Peter Thiel's secretive 'Dialogue' society, which hosts sessions on topics from cult-building to nuclear policy.
  • Midjourney, a bootstrapped AI image company, is developing a consumer body scanner it claims will replace MRIs, funded entirely by its $200M+ annual revenue from image generation.
  • A SemiAnalysis study found AI subscription plans are heavily subsidized, with OpenAI's $200/month ChatGPT Pro offering $14,000 in token value and Anthropic's $200/month Claude Max providing $8,000 worth.
  • The hosts warn that autonomous drone swarms represent a fundamental shift in warfare, being cheap, asymmetrical, and difficult to counter with traditional jamming or small arms.
  • Marty and Matt advise against locking business operations into a single AI provider like Claude Code, recommending agentic harnesses that allow easy model switching to avoid vendor lock-in and regulatory risk.
Also from this episode: (3)

Protocol (3)

  • The hosts argue the new Illinois Digital Asset Tax Act is a predatory law designed to criminalize financial privacy, not generate revenue, by imposing a 0.2% tax on all crypto transactions.
  • Marty argues Bitcoiners in Illinois should consider moving their families and businesses to friendlier jurisdictions, citing his own exit from New York as a precedent for voting with your feet.
  • A bug in Bitcoin Core v31.0’s private broadcast feature can leak a node's IP address if a V2 transport handshake fails, compromising privacy for users not routing through Tor.
#Safety#Regulation#Agents#Privacy#Coding
The AI Daily Brief: Artificial Intelligence News and Analysis
The AI Daily Brief: Artificial Intelligence News and Analysis

Nathaniel Whittemore

The Models Trying to Fill the Fable Gap • Jun 18

  • Nathaniel Whittemore reports the US government ordered Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national worldwide citing national security concerns, a directive Anthropic received at 5:21 p.m. Eastern.
  • The Wall Street Journal reported Commerce Secretary Howard Lutnik issued the export restriction letter to Anthropic CEO Dario Amade. Journalist Prince cited Axios reporting that the government unsuccessfully asked Anthropic to pause the model release.
  • Anthropic argued in its blog post the government's national security concern was a narrow, non-universal jailbreak discovered by researchers, likely at Amazon under Project Glasswing. Anthropic claims this method finds vulnerabilities already widely discoverable via other models.
  • Anthropic stated that perfect jailbreak resistance is impossible for any model provider and defended its defense-in-depth strategy, including 30-day customer data retention to research and mitigate jailbreaks.
  • AI entrepreneur Bindu Ready argued the government's pretext is weak because every model can be prompted to reveal common security vulnerabilities or other sensitive information that is freely available on the internet.
  • Nathaniel Whittemore notes the policy's immediate impact blocks key Anthropic technical staff like Andre Carpathy, who are non-US citizens on visas, from accessing the company's own models.
  • CFR senior fellow Chris Magcguire criticized the Department of Commerce export control strategy as incoherent, simultaneously sending advanced AI chips to China while banning model access globally.
  • Nathaniel Whittemore highlights the perceived hypocrisy: the White House Office of Science and Technology Policy recently defended voluntary model sharing as not being government oversight, citing chilling effects on free speech and innovation.
  • AI builder Sarah Hooker and Jeremy Howard criticized Anthropic's arrogance in pursuing the Fable release, assuming their unique safety stance would be accepted.
  • Will Manitus cited Anthropic's earlier blog post advocating for government power to block unsafe model deployments, framing the current export control as a 'say-around, find-out' moment for the company.
  • Entrepreneur John Enis argued Anthropic's safety song-and-dance was largely due to compute constraints for serving Mythos at scale, not genuine danger, and that launching Fable with restrictions was an IPO strategy.
  • Nathaniel Whittemore reports Eleazar Yudkowski's nuanced reaction, stating international treaties to halt AI escalation would be good, but the current action's selectivity and overrule aspects are bad.
  • Aaron Levy stated this action creates a precedent where governments deem some models too powerful for certain uses, marking a major turning point for AI regulation.
  • Sterling Crispen coined 'capability thought crimes,' warning the precedent could require DOW clearance for future model releases and be detrimental for progress.
  • Brian Xiao outlined operational impacts: US companies will need ID verification for API billing, affecting downstream services like Cursor and Harvey. Frontier labs like OpenAI and Google DeepMind now lack incentives to release mythos-caliber models.
  • Connor Brown compared this to the 1990s cryptography wars, predicting a fight over KYC and anti-compete laundering laws for frontier models with much higher stakes.
  • Daniel Woo argued this precedent torpedoes the AI bull case by restricting monetization of powerful models and creates compliance headaches for US enterprises with non-US national employees.
  • Nathaniel Whittemore posits the entire US economy rests on Anthropic and OpenAI's revenue growth, and the government's move damages the whole economy's AI buildout.
  • VC Hemtt Mahabra and Alex Petropolis warned this triggers a sovereign AI moment, where nation states will require citizenship or security clearances for state-of-the-art model work, forcing middle powers to build leverage.
  • Gail Weiner argued the US narrative of being a predictable, rule-of-law provider evaporated, giving procurement officers globally defensible arguments for sovereign AI hedging or experimenting with Chinese openweight alternatives.
  • Nathaniel Whittemore cites the 'Europeans account' tweet framing this as technological dependence: European entities built on frontier models could see workflows stop overnight, costing millions in emergency replacements.
  • Malon X described the dystopian outcome as a caste system based on access to frontier intelligence, a new digital iron curtain dividing those allowed to accelerate science from citizens of other countries.
#Models#Regulation#Safety#Big Tech#Startups

A Big Shift in the AI Race • Jun 17

  • Anthropic sent a technical team to negotiate with Washington, including Chief Compute Officer Tom Brown and security researcher Nicholas Carlini, whose March demonstrations convinced him AI models are superior vulnerability researchers.
  • Commerce Secretary Howard Lutnick phoned into the meeting from the G7 summit, but key decision-makers like National Cyber Director Shawn Carrencross were absent, suggesting negotiators lacked authority to lift the ban.
  • The government's primary concern is Mythos, not Fable. Commerce expressed willingness to restore Fable if Anthropic fixed the jailbreak, but fears Chinese access via a South Korean telecom added to Mythos's Project Glasswing.
  • Leda Security CEO Katie Moussouris argues the Fable jailbreak behavior - refusing to review insecure code but generating patches - is standard cyber defense, shared by GPT-5.5 and Opus-4.8.
  • Administration officials claim Anthropic knowingly distributed a jailbreak-susceptible model and failed to communicate, with one source stating, 'They screwed us.'
  • Charlie Bullock of the Institute for Law and AI says the government's legal theory is aggressive and vulnerable to challenge, but Anthropic will likely settle without litigation.
  • Ashley (@paulaulithia) frames the conflict as Anthropic failing regulatory protocol: they scoped risk too broadly, then narrowed it when mitigations proved insufficient, triggering enforcement.
  • Whittemore argues Anthropic has not realized managing its US government relationship is now as critical as building models, citing delayed responses to official contacts.
  • Over 100 cybersecurity experts signed an open letter stating removing Mythos from defense tools increases vulnerability. Agathe Demarais wrote the ban boosts Chinese AI appeal.
  • Deutsche Bank's Jim Reid warned a prolonged ban threatens US tech firms and AI adoption speed, as you cannot rely on technology that could be switched off.
  • Helen Toner stated it's widely agreed you cannot fully fix jailbreaks in these models, making the government's demand for a patch potentially unfeasible.
  • SpaceX's IPO closed at $201.80, up 49% from its IPO price, giving the company a $2.6 trillion valuation and making it the world's fifth largest company.
  • Elon Musk became the world's first trillionaire post-IPO, holding a 46% stake in SpaceX. Flexport CEO Ryan Petersen noted Musk's single-day gain exceeded Warren Buffett's lifetime earnings.
  • SpaceX acquired Cursor for $60 billion; Cursor had a $4 billion run rate and was growing 7x year-over-year. Its Composer 2.5 model matches Opus 4.7/GPT 5.5 performance at a tenth of the cost.
  • Cursor engineer Nick Dobos teased a new from-scratch model, same size as Opus and GPT 5.5, using 10-20x more compute than Composer, aimed for general intelligence.
  • The DOJ intervened in an NAACP lawsuit against xAI, arguing Grok is vital to national security and supports Department of War missions, including recent strikes on Iran.
Also from this episode: (2)

Big Tech (2)

  • Ed Zitron published OpenAI's audited numbers: a $5B net loss in 2024 and a $38.5B net loss in 2025, but OpenAI claims $30B of the 2025 loss was a non-cash accounting change from restructuring.
  • OpenAI's inference business shows profit margins: $3.7B revenue on $2.7B cost in 2024, and $13B revenue on $7.5B cost in 2025. The company holds $73B in cash and securities.
#Big Tech#AI Infrastructure#Startups#Enterprise

The Fable 5 Crisis Continues • Jun 15

  • The US government ordered Anthropic to suspend all foreign access to Fable 5 and Mythos 5, citing a national security concern from a reported jailbreak. The directive came via Commerce Secretary Howard Lutnik.
  • Anthropic says the government’s letter cited a specific jailbreak technique, viewing it as bypassing safeguards. The company reviewed the technique and found it only identified minor, previously known vulnerabilities.
  • Anthropic argues perfect jailbreak resistance is impossible today. Their defense-in-depth strategy for Fable 5 aimed to make jailbreaks narrow or expensive, combined with monitoring and a 30-day customer data retention policy.
  • Anthropic contends the jailbreak capability displayed is widely available from other models, including OpenAI's GPT 5.5. They believe applying this standard across the industry would halt all new frontier model deployments.
  • Wall Street Journal reported the jailbreak research was conducted by Amazon. The government contacted Anthropic to pause the model release beforehand, but Anthropic refused.
  • Katy Mures of Letter Security and AI policy expert Dean Ball called the government's action a complete overreaction and cartoonish, arguing the prompting used is exactly what defenders do.
  • Many observers felt the policy was hypocritical and personal, citing White House statements weeks earlier that argued government overreach would chill free speech and innovation.
  • John Enis argued Mythos is the best model but not world-changingly dangerous. He claimed Anthropic's safety rhetoric was for regulatory capture and compute scarcity, and their cynical plan backfired.
  • Aaron Levy and Andrew Friedman described this as a major turning point for AI regulation, creating precedent for government controls on model access and intervention in capitalist society.
  • Connor Brown framed this as the start of AI wars, comparable to the 1990s cryptography fight but with higher stakes. He predicted KYC and anti-compete laundering laws for frontier models.
  • Malon X described the move as creating a dystopian caste system based on access to frontier intelligence, a new digital iron curtain separating citizens by allowed cognitive acceleration.
Also from this episode: (6)

Enterprise (4)

  • The export control order impacts foreign nationals within Anthropic, including Andre Carpathy and others on EB1 visas, prohibiting them from interacting with their own models.
  • Industry scorn targeted Anthropic for its perceived arrogance and safety scaremongering. Critics point to Anthropic’s own blog post advocating government power to block unsafe deployments, framing this as a 'fear around, find out' moment.
  • Brian Xiao outlined practical impacts: American companies will need citizenship verification for API access, affecting downstream services like Cursor and Harvey. Frontier labs lose incentives to release mythos-caliber models.
  • Global observers warned of sovereign AI fragmentation. Gail Weiner argued the US's predictable rule-of-law narrative evaporated, empowering procurement officers in Brussels or Tokyo to seek sovereign AI hedging.

AI Infrastructure (1)

  • Chris Magcguire criticized the Commerce Department's incoherent export control strategy, highlighting it sends chips to China, fails to enforce existing controls, and now prevents US AI companies from releasing their own models.

Markets (1)

  • Market analysts warned the precedent could torpedo the AI bull case, as labs face rising capex but restricted monetization. The move injects permanent risk into AI investing and threatens Anthropic's IPO valuation.
#Safety#Regulation#Trade#Big Tech

Fable 5 Shut Down by US Government • Jun 13

  • The US government ordered Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals, citing a discovered jailbreak method as a national security threat.
  • Anthropic argues the directive is based on a narrow, non-universal jailbreak that reveals vulnerabilities also found in other public models like GPT-5.5, not a unique threat.
  • Nathaniel Whittemore notes the government's action creates precedent for regulating model capabilities and could stifle progress by making companies like OpenAI and DeepMind hesitant to release advanced models.
  • Critics like Dean Ball and Emerson Brooking view the directive as incoherent or hypocritical, given the administration's stance on exporting AI chips to China while restricting model access.
  • Industry sentiment blames Anthropic for provoking regulation with its fear-mongering about model dangers, referencing Dario Amodei's earlier blog post advocating for government blocking power.
  • Nathaniel Whittemore highlights the operational impact: API providers must verify user citizenship, internal researchers like Andrej Karpathy cannot access models, and global workflows dependent on Fable are disrupted.
  • Observers like Aaron Levy and Andrew Friedman see this as a major turning point, signaling government intervention in AI progress and moving towards capability-based regulation.
  • The move prompts fears of a digital Iron curtain, where access to frontier intelligence becomes citizenship-based, and non-US entities pursue sovereign AI development.
  • Nathaniel Whittemore argues the directive threatens the US economy by undermining investor confidence in AI capex and Anthropic's IPO, while empowering open-source and foreign alternatives.
  • Eliezer Yudkowsky acknowledges the action's complexity, noting it could push against unchecked model escalation but also create selective overreach, making the long-term outcome unclear.
#Safety#Regulation#Big Tech#China
Breaking Points with Krystal and Saagar
Breaking Points with Krystal and Saagar

6/15/26: Anthropic Model Banned, AI Crash Incoming, Eric Trump UFC Scandal, WH Panic Over Situation Room Leak • Jun 15

  • Krystal argues the government's selective enforcement against Anthropic reflects political favoritism, citing Dario Amodei's disfavored status versus Sam Altman or Elon Musk's closer ties to the Trump administration.
  • Saagar argues AI development needs rigorous oversight akin to drug approval, not ad-hoc crackdowns, citing the disparity between billions in FDA testing and zero scrutiny for frontier AI.
  • Anthropic had previously spread its Mythos model to 150 organizations across 15 countries before the public release of its successor Fable Five.
  • Ed Zitron argues LLMs like Fable Five are oversold hype, inherently limited by hallucinations and unsustainable costs, and do not replace human labor.
  • Zitron claims AI companies are financially unsustainable, with training costs a permanent burden, and market valuations depend on semiconductor hype not real revenue.
  • Zitron predicts an AI bubble collapse will trigger a fundamental revaluation of the tech industry and a permanent breakdown of venture capital, citing the sector's lack of new hypergrowth ideas.
  • Fighter Josh Hokit shouted 'Michelle Obama is a man' in his post-fight interview at the White House UFC event, repeating a provocation he used on the Contender Series.
  • Luke Thomas argues the UFC has been the most important mainstream actor rehabilitating Trump's image post-January 6th, providing him third-party validation and direct access to young male voters.
  • UFC bonuses for fighters at the White House event were paid by World Liberty Financial, a company central to Trump's financial self-dealing.
  • Polling shows only 16% of Americans and 31% of Republicans approve of hosting UFC fights on the White House lawn.
  • The White House fears Maggie Haberman and Jonathan Swan's book contains verbatim Situation Room quotes from tapes, as no official has disputed the exact dialogue.
  • Swan's book reveals JD Vance privately opposed the Iran war but publicly supported Trump, while Dan Bongino screamed at the Chief of Staff over the Epstein cover-up strategy.
Also from this episode: (3)

Enterprise (2)

  • The US government issued an export control directive suspending all foreign access to Anthropic's Fable Five and Mythos AI models citing national security concerns.
  • Anthropic claims Amazon CEO Andy Jassy alerted US officials after researchers at Amazon jail-broke Fable Five, prompting the government crackdown.

Sports (1)

  • Daniel Cormier tweeted screenshots alleging Eric Trump asked him about rigged fights and fighter injuries ahead of UFC 250, which Cormier later deleted.
#Big Tech#Regulation#Markets#Media

6/15/26: Trump Says Iran Deal Is Done, Jeremy Scahill On US Capitulation, Israelis Meltdown Over Deal • Jun 15

  • Israeli political opposition is unified: Yair Lapid calls the deal Israel's greatest strategic failure, and Bezalel Smotrich vows to continue a campaign to topple Iran in 'creative ways.'
  • Saagar cites Israeli Telegram comments framing the war as a loss and Iran teaching America a lesson, while US pundits like Mark Levin warn the deal emboldens Iran to attack Israel.
  • Lindsey Graham's statement positions JD Vance as the deal's architect, signaling future congressional scrutiny and potential criticism without directly attacking Trump.
  • Iran argues Gaza requires a separate diplomatic track and lacks the same leverage as Lebanon, where Hezbollah directly joined the warfront, dispelling the myth of Iran as Hamas's puppet master.
  • Scahill notes Israel has killed approximately 1,000 Palestinians in Gaza since the October ceasefire, exceeding the number of Israeli civilians killed on October 7th.
Also from this episode: (5)

War (3)

  • Jeremy Scahill reports Trump faced imminent Iranian missile retaliation after Israel bombed Beirut, forcing last-minute concessions including a more rapid Strait of Hormuz reopening and a US pledge to compel Israeli withdrawal from Lebanon.
  • Scahill describes Trump's deal as capitulation, returning to pre-war leverage points but with Iran stronger, and argues the final terms may be worse than Iran's pre-war February offers.
  • Scahill says Iranian officials anticipate another war with Israel, viewing the current deal as a phase, and believe the US-Israel strategy aims to decouple Hezbollah from Iran.

Iran (1)

  • Iran's published deal terms require $24 billion in frozen assets released, $300 billion in reconstruction plans, full sanctions suspension, and removal of its missile program and support for resistance groups from the agenda.

Diplomacy (1)

  • Scahill notes the deal's provisional structure gives Iran leverage: if the US fails to enforce terms like a Lebanese ceasefire, Iran can halt the 60-day nuclear negotiation window.
#Diplomacy#War#Middle East

Related Stories