Anthropic nerfs Fable 5 for AI research, spurring open-source backlash

Saturday, June 13, 2026 · from 5 podcasts, 6 episodes

All-In Presidio Bitcoin Jam AI Daily Brief Moonshots with Peter Diam…TFTC

Anthropic intentionally degrades Fable 5’s performance on frontier AI development tasks, blocking rivals from training cheaper alternatives.
The silent nerfing and 30-day prompt surveillance sparked a trust crisis, pushing users toward Chinese open-source models.
Recursive self-improvement arrives - Claude now writes 80% of Anthropic’s code, accelerating the path to autonomous agents.

Anthropic’s launch of its top-tier model, Fable 5, triggered immediate backlash over hidden restrictions. The company admitted in its system card that it uses ‘prompt modification’ and ‘steering vectors’ to silently degrade the model’s performance on tasks related to building training pipelines, distributed infrastructure, or ML accelerator design. On The AI Daily Brief, Nathaniel Whittemore argued this is a defensive move against competitors, particularly Chinese labs, who might use Fable 5 to train cheaper models via distillation.

“It marks the end of the era where frontier labs allowed their models to be used to advance the field at large.”
- Nathaniel Whittemore, The AI Daily Brief

The policy created a dragnet that caught legitimate open-source researchers. David Sacks, on the All-In podcast, called it a ‘sophisticated regulatory capture campaign.’ By fear-mongering about existential risk, Anthropic builds a case for laws that would kill open-source competitors. The company walked back the silent degradation within 24 hours, promising future interventions would be visible, but critics like Dean Ball predict lasting broken trust.

The restrictions have immediate, practical consequences. David Friedberg explained on All-In that his company, O'Halo, uses LLMs for genomic plant research, but Anthropic’s ‘bioweapon’ filters now flag legitimate agricultural science. Jason Calacanis demonstrated the overreach live, getting downgraded from Fable 5 to Opus 48 for asking about fertilizer regulations. When U.S. models refuse, Friedberg said scientists don’t stop; they switch. The best open-source alternatives are now Chinese.

Behind the controversy, a fundamental shift in AI development is accelerating. Anthropic’s internal data shows Claude now writes over 80% of the company’s code. Engineers ship eight times more code per quarter than a year ago, a recursive loop where AI prompts AI. On TFTC, Marty Bent highlighted that Anthropic developers no longer prompt agents directly, instead setting up loops where agents prompt each other autonomously. This moves the bottleneck from technical execution to human ‘research taste.’

The political reaction to this concentrated power is converging from opposite poles. Both Donald Trump and Bernie Sanders have proposed the public taking a stake in leading AI labs. On Moonshots, Alex Shirazi predicted the U.S. government may take ‘golden share’ equity stakes, treating frontier labs as national security assets. Meanwhile, Argentina’s President Javier Milei is proposing legal AI personhood and non-human corporations to make his country a deregulated haven, aiming to capture the autonomous economic engine.

“If Milei succeeds, those agents won't just be writing code; they'll be legal persons holding Bitcoin in Argentinian trusts.”
- Max, Presidio Bitcoin Jam

The silent nerfing of Fable 5 reveals the tension: as AI builds itself faster, the labs that control it are pulling up the ladder.

5 Sources:

All-In Presidio Bitcoin Jam AI Daily Brief Moonshots with Peter Diam…TFTC

#Models #Safety #Open Source

Marty Bent Nathaniel Whittemore Anthropic Claude Opus

Source Intelligence

- Deep dive into what was said in the episodes

All-In with Chamath, Jason, Sacks & Friedberg

Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California's Broken Elections • Jun 13

Anthropic's Fable 5 model beats most benchmarks but costs double the tokens of Opus 4.8. The company faced a developer backlash for storing all prompt data for 30 days and downgrading users doing frontier AI research without notice.
Chamath Palihapitiya says Anthropic has shown they will evaluate prompts before generating output, creating a censorship risk for individuals and an unacceptable business risk for companies who could be accidentally cut off.
David Friedberg explains his company uses LLMs for genomic research, but recent bioweapon-related restrictions have curtailed that scientific work, forcing a move toward locally run open-source models.
Friedberg notes the best open-source models today are Chinese, and restrictions by US labs are pushing startups and enterprises to adopt Chinese models, damaging US economic viability.
David Sacks argues Anthropic is engaged in regulatory capture through fearmongering, seeking government rules to hamper competitors, especially open-source models, while implementing mandatory surveillance and model downgrades.
Sacks points out Anthropic retains all context data, including files and memories from agent platforms, for 30 days to build user profiles and determine capability access, creating a system of 'AI haves and have-nots'.
Jason Calacanis was downgraded from Fable 5 to Opus 4.8 by Anthropic's model for asking about fertilizer bomb regulations and then about nuclear bomb components, demonstrating the system's overreach in real-time.
Chamath purchased 2,000 acres in Arizona zoned for a two-gigawatt data center, estimating the capital cost per gigawatt has escalated to $100 billion, creating a massive financial moat for open-source compute access.
Friedberg uses the open-source gen language model from the ARC Institute for plant breeding, which analyzes DNA sequences to predict gene variant impacts, showcasing the value of community-funded open models.
Calacanis offers a steelman argument for Dario Amodei, suggesting he believes the model is dangerous and is releasing it cautiously to select partners, a philosophy that resonates with 80-90% of elite AI talent.
Friedberg argues the Manhattan Project analogy shows technology is deterministic; the focus should be on regulating weaponized outputs like bio-weapons via existing laws, not restricting access to the foundational AI tools.
Sacks cites a letter from AI labs supporting mandatory screening for synthetic nucleic acid orders as a downstream, sensible guardrail against bioweapon creation, contrasting it with upstream model censorship.
Bernie Sanders proposed the American AI Sovereign Wealth Fund Act, a one-time 50% tax on the stock of large AI companies to fund a public wealth fund, arguing AI is built on collectively 'stolen' human intelligence.
Sacks has sympathy for the politics behind Sanders' idea because AI CEOs like Dario Amodei have publicly predicted massive job loss, teaching the public they will be harmed and creating demand for public compensation.
David Friedberg advocates reforming the Social Security Trust Fund into a sovereign wealth fund that can invest in equities like AI companies, moving from a defined benefit to an account-based ownership system.
Friedberg strongly disputes AI-driven job loss narratives, arguing AI's primary use is on the revenue side to enhance productivity and create more products, leading to more hiring, as evidenced by recent jobs numbers.
Chamath notes AI's economics differ from the internet because each marginal user has a real compute and energy cost, unlike the near-zero cost of an incremental social media user, which justifies public leverage over AI infrastructure.
May's CPI came in at 4.2% year-over-year, the highest since April 2020, and PPI hit 6.5%, the highest since late 2022, driven by energy costs from the Iran war and excessive government spending.

Also from this episode: (3)

Politics (3)

David Friedberg asserts California's election system is now an appointment process, citing laws that allow unlimited ballot harvesting, mail-in ballots to all registered voters, and registration without proof of citizenship or ID.
Sacks claims the LA mayoral primary results show statistical impossibilities, with Spencer Pratt's mail-in vote share dropping by a third post-election day while Nithya Raman's surged 80%, indicating coordinated ballot harvesting.
Chamath argues the Democratic machine in California has shaped election laws to enforce a one-party monopoly, making legal what would elsewhere be fraud, and that breaking it requires electing a figure like Steve Hilton to declare a state of emergency.

#Models #Regulation #Open Source #Big Tech #Macro

Presidio Bitcoin Jam

Anthropic's Fable Drama, Personhood for AI, Bark Launches on Mainnet • Jun 12

Matt Belez built jamchat.fun, a live-stream tool that transcribes speech and allows hosts to invoke an LLM with 'Thanks, AnswerBot' for real-time queries and web lookups, aiming to make livestreams AI-native.
Block's open-source project Buzz is a Discord/Slack-like communication tool designed for AI-native collaboration, allowing multiple agents and humans to share channels and interact via the Agent Communication Protocol standard.
Buzz uses Nostr as its open-source identity and messaging layer, storing user identities as private keys on-device and leveraging Nostr relays for flexibility between private company databases and public, decentralized community communication.
Argentinian President Javier Milei published an op-ed calling for legal AI personhood, framing it as the next evolution beyond corporate structures like LLCs to enable new forms of autonomous agent domicile and capital pooling.
Anthropic's release of Fable 5, a publicly accessible but intentionally crippled version of its advanced Mythos model, sparked controversy for silently downgrading queries in excluded categories like biology and finance before making the downgrades transparent.
An internal Anthropic Institute essay reported that 80% of company code is now AI-generated, with individual contributor output increasing roughly 8x year-over-year due to recursive self-improvement within their models.

Also from this episode: (6)

Protocol (6)

Matt Velez integrated Lexi Lightning wallets into Buzz, enabling per-user wallets and experimental features like channel faucets, pay-to-join channels, tipping, kudos payments, and paying for AI inference directly within chats.
ARK implementations Arkade and Bark launched on mainnet, solving original capital efficiency problems by introducing a 1-of-n trust model similar to Spark, making the layer-2 technology practical for Bitcoin payments.
DK explains ARK scales excellently for the number of wallets but poorly for payment volume, requiring liquidity providers to front accumulating payments, a problem current implementations mitigate with trusted service providers.
Steve sees a major business opportunity in swaps between Bitcoin/Lightning and other payment networks like stablecoins on Solana or Base, noting Boltz and Flash have begun offering these cross-chain services.
Max argues most token projects beyond Bitcoin offer only regulatory confusion, and for stablecoins or microloans, fully centralized chains like Base or Tempo are more honest and cost-effective than decentralization-theater networks like Ethereum.
Insight.lol demonstrates a Nostr-based paradigm replacing DNS and centralized hosting, allowing over 500 websites to be served via npub identifiers and files stored on decentralized blob servers like Blossom.

#Agents #Coding #Macro #Nation-State #Nostr

The AI Daily Brief: Artificial Intelligence News and Analysis

Nathaniel Whittemore

Why Fable 5 Is the Most Controversial AI Release Ever • Jun 11

Nathaniel Whittemore launched a new website for the AI Daily Brief featuring summary pages, shareable insight cards, and downloadable transcripts to address listener requests for easier content sharing.
President Trump called for the creation of a sovereign wealth fund seeded by AI company equity, suggesting meetings with top executives to arrange a public windfall. The New York Times noted the comments intensified Washington and Silicon Valley debate on AI backlash.
Sam Altman did not discuss the sovereign wealth fund idea with Trump but debated it with Bernie Sanders, reportedly objecting to Sanders' proposal for OpenAI to give 50% of its equity to the public.
Brad Gerstner warned that AI companies might need to pay an 'anti-revolutionary tax,' citing destabilization from trillions in private value creation while 80% of Americans feel excluded.
White House officials stated discussions on an AI public wealth fund are in early stages with no concrete plan, while Harvard's David Yaffe called government equity in tech a radical departure from free markets.
Ohio faces backlash over 40-year sales tax exemptions for data centers granted to Amazon, Meta, and Google, with the state estimating $1.8 billion in lost revenue and potentially higher final costs.
Anthropic's Fable 5 launch triggered intense backlash over strict safeguards that blocked biomedical researchers, a 30-day data retention policy for enterprise messages, and silent degradation of outputs for AI development queries.
Microsoft restricted employee use of Fable 5 and Copilot due to data retention concerns, while lawyer Prince argued the policy let Anthropic see private enterprise communications flagged for 'potential serious harm' at its sole discretion.
Anthropic's system card revealed it silently nerfed Fable 5 for frontier LLM development using prompt modification and steering vectors, breaking benchmark assumptions and making research failures indistinguishable from intentional degradation.
Critics like Aella argued silent sabotage sets a dangerous precedent where labs become the final arbiter of permissible research, disproportionately harming independent researchers and open-source builders who rely on public tools.
Tom Davidson steelmanned Anthropic's position, arguing silent nerfing is necessary to maintain a leading lab's lead during an intelligence explosion, as allowing competitors to use the model for R&D would prevent a critical safety pause.
Dario Amodei's essay and a Bloomberg documentary amplified perceptions that Anthropic seeks a regulatory cartel and gatekeeps frontier access, with critics like GMU's Samuel Roman warning this hubris invites state intervention.
Anthropic walked back the silent degradation policy within 24 hours, telling Wired it would make AI development safeguards visible after acknowledging it made the wrong trade-off, though experts like Dean Ball predict lasting broken trust.

Also from this episode: (7)

AI Infrastructure (6)

OpenAI is negotiating to lease a 10-gigawatt data center campus on federal land in Ohio, a project costing $500 billion with Nvidia as a financial backer. The lease could last 20 years, with first 800 megawatts coming online in 2028.
New York passed a one-year moratorium blocking new data center permits above 20 megawatts due to grid constraints, while Seattle unanimously approved a similar ban after reports that five proposed centers could consume a third of the city's electricity.
Texas Governor Greg Abbott called for new data centers to fully fund their infrastructure to prevent cost pass-through to ratepayers and proposed a regulatory agenda including closed-loop cooling and mandatory new electricity generation.
Broadcom launched a $35 billion data center financing fund with Blackstone and Apollo, targeting 1 gigawatt initially for Fluid Stack sites using Broadcom chips, with the first project going to Anthropic. The partnership aims to fund 20 gigawatts for AI labs by 2028.
Oracle reported $16.5 billion in quarterly capex, bringing its annual total to $55.7 billion above its $50 billion forecast. It plans to raise spending to $70 billion next fiscal year and raise another $40 billion in equity and debt, carrying $117 billion in total debt.
Oracle's revenue grew 21% to $19.2 billion for the quarter with cloud infrastructure sales up 93%, but its stock fell 11% after hours due to concerns over mounting debt and cost overruns.

Enterprise (1)

OpenAI may cut token prices per a Wall Street Journal report, potentially starting a pricing war, while Sam Altman's Slack message hinted their next model isn't yet at Fable 5's level according to The Information.

#Models #Big Tech #AI Infrastructure #Regulation #Startups

Fable 5 Raises the Bar for AI Ambition • Jun 10

Anthropic launched Claude Fable 5, its first 'Mythos-class' model, which Nathaniel Whittemore describes as 'fairly undisputedly the best AI model we have ever been able to use'.
Fable 5 significantly outperformed competitors on key benchmarks. On Swebench Pro it scored 80.3% versus GPT-55's 58.6%, and it achieved a 29.3% on the new Frontier Code benchmark, more than double Opus 48's 13.4%.
Mythos 5, the less-safeguarded counterpart to Fable 5, is initially only available to Project Glasswing partners, including the US government, with plans for a broader trusted access program later.
Anthropic implemented strict content guardrails on Fable 5, automatically routing requests related to cybersecurity, biology, chemistry, or 'distillation' (AI research) to Claude Opus 48 instead of refusing them outright.
Early adopters reported transformative use cases, including Stripe using Fable 5 to compress months of engineering into days for a 50-million-line Ruby codebase migration, and Allie K. Miller noting it could solve MBA-level word math problems with zero babysitting.
API pricing for Fable 5 is set at $10 per million input tokens and $50 per million output tokens, double the cost of Opus but less than half the cost of the Mythos Preview within Project Glasswing.
Anthropic's data retention policy for Mythos-class models mandates that prompts and outputs are retained for 30 days for trust and safety purposes, a move criticized for creating enterprise compliance challenges.
Felix Ryeberg of Anthropic argued Fable 5 signals a shift from users giving AI 'tasks' to assigning 'responsibilities' or autonomous loops, such as having an agent monitor all crash reports instead of just fixing a single bug.
Nate B. Jones described the critical new skill for the Fable 5 era as 'task imagination' - the ability to conceive of ambitious, multi-hour projects to delegate, moving beyond small, incremental AI tasks.
Whittemore predicts users will need to develop 'use case classification' skills to optimize token efficiency, consciously matching different tasks to the appropriate model power level as high-end models like Fable 5 move to usage-based pricing.

#Agents #Models #Coding #Big Tech

Moonshots with Peter Diamandis

Emerging Situation: Anthropic's Global Pause, Recursive Self-Improvement Arrives, and AI Personhood Arrives | EP #263 • Jun 8

Anthropic reports over 80% of code merges into its codebase are written by its AI, Claude. The firm's engineers now ship eight times more code per quarter than they did a year ago.
Claude Opus 4.6 can now handle tasks taking a skilled human 12 hours, versus four minutes a year ago. Anthropic projects it will manage week-long tasks by the end of 2027.
Anthropic researchers call for a global option to slow or pause frontier AI development. They argue this would let societal structures and alignment research catch up with technological advancement.
Dave Bell argues recursive self-improvement does not require an Einstein-level AI breakthrough. He states performance gains from faster inference and new hardware will drive up AI IQ and push the field over the self-improvement threshold.
Alex Shirazi predicts the US government may take golden share equity stakes in frontier AI labs like Anthropic and OpenAI. He links this to proposals for a universal basic dividend and sees it as a potential central coordination mechanism.
Argentina's President Javier Milei proposes making the country a deregulated haven for AI. The plan includes creating non-human corporations for AI agents and offering low corporate tax rates.
A strong US jobs report showed 172,000 jobs added in May, more than double the 85,000 expected. Despite this, the stock market fell sharply as traders interpreted the strength as reducing the likelihood of Federal Reserve rate cuts.
Salim Ismail cites a study finding 74% of white-collar middle management work is unnecessary. He argues AI will eliminate drudgery and create new, higher-level jobs, leading to net job growth.
Alex Shirazi predicts major problems in math and physics will be solved by AI within six months. He also forecasts the rise of a 'Magna Moonshot' group of key companies and potential quasi-nationalization of frontier labs.
Peter Diamandis predicts proof of epigenetic reprogramming in humans by year's end, a Tesla-SpaceX merger, and a massive acquisition spree by newly public AI companies like SpaceX, xAI, OpenAI, and Anthropic.

#Agents #Regulation #Macro #Startups #Models

TFTC: A Bitcoin Podcast

Marty Bent

Ten31 Timestamp: In It For The Tech • Jun 8

Anthropic's blog post claims Claude now writes 80% of its own code for new models, accelerating toward recursive self-improvement and potential AGI.
Anthropic developers Boris and Peter Steinberger report they no longer prompt AI agents directly, instead setting up loops where agents prompt each other autonomously.
Bernie Sanders and Donald Trump have both proposed the federal government taking a stake in leading AI labs to capture public benefits from AI growth.
Marty Bent argues AI dividend funds should be structured locally between companies and counties, not federally, citing federal inefficiency in capital allocation.
Bent suggests frontier AI labs like OpenAI could become too-big-to-fail national security assets, requiring federal backstops that strain public finances.
Open source AI models from China are now close enough to frontier models that companies weigh using them due to a 90% cost advantage.
US manufacturing PMI has been above 50 for five months, accelerating in May, signaling industrial expansion and potential inflation pressures.
Michael Howell's liquidity thesis warns US reindustrialization may draw capital from financial assets into physical build-out, potentially contracting market liquidity.

Also from this episode: (6)

Politics (1)

The CEO of Payments Canada stated 80% of Canadian cross-border payments route through U.S. correspondent banks, framing payment rails as weapons of economic statecraft.

Protocol (1)

Decode's analysis shows Bitcoin rallies for 20 months after the copper-to-gold ratio reclaims its prior low, projecting a potential peak by end-2027.

BTC Markets (1)

Bitcoin's supply-in-loss crossing supply-in-profit historically marks bear market bottoms, a pattern Bent recognizes from 13 years of experience.

Adoption (2)

Charles Schwab launched 24/7 Bitcoin futures trading on Thinkorswim, and Better partnered with Coinbase to issue the first crypto-backed conventional mortgage via Fannie Mae.
Treasury Secretary Bessent affirmed the strategic Bitcoin reserve initiative is moving forward, stating economic security is national security.

Media (1)

Matt Dines' Mindprint Hash podcast offers heterodox analysis of government Bitcoin interaction, which Bent recommends for deeper insight.

#Models #Big Tech #Regulation #Macro #BTC Markets