Cory Doctorow critiques Milton Friedman's shareholder supremacy doctrine, arguing that the standard of "maximizing shareholder value" is actually unfalsifiable and subjective, not the objective bright-line test Friedman claimed. He contends that the ambiguity of this standard serves as a convenient excuse for corporate leaders to justify cruelty and cowardice under the guise of fiduciary duty.
Cory Doctorow critiques Milton Friedman's shareholder supremacy doctrine, arguing that the standard of "maximizing shareholder value" is actually unfalsifiable and subjective, not the objective bright-line test Friedman claimed. He contends that the ambiguity of this standard serves as a convenient excuse for corporate leaders to justify cruelty and cowardice under the guise of fiduciary duty.
Gary Marcus argues that OpenAI's competitive advantage in AI is fading due to a lack of a protective "moat," as rivals increasingly match or surpass its technology and open-source alternatives gain traction.
A German court ruled that Google can be held liable for defamatory fake information generated by its AI search tool. The case involved a chatbot that falsely claimed a person had made insolvency-related statements. This decision sets a precedent for AI-generated content liability in Germany.
Leaked financial documents reveal that OpenAI is losing billions of dollars annually, spending far more on computing, staffing, and operations than it generates in revenue through products like ChatGPT and API services.
This analysis was generated by AI and may contain inaccuracies. Always verify with original sources.
On June 13, 2026, the US government issued an export control directive under national security authorities ordering Anthropic to suspend all access to its Fable 5 and Mythos 5 models for any foreign national, both inside and outside the United States, including foreign national employees of Anthropic itself. The directive was received at 5:21pm ET and gave no specific details about the underlying national security concern. The government's stated rationale centers on a purported jailbreak technique that allegedly enables bypassing the models' safety guardrails.
Anthropic's public statement explains that the government provided only verbal evidence of a potential "narrow, non-universal jailbreak" that essentially consists of asking the model to read a specific codebase and fix any software flaws. Anthropic states that this level of capability is already widely available from other models, including OpenAI's GPT-5.5, and is routinely used by security defenders. The company noted that the specific technique could identify a small number of previously known, minor vulnerabilities, and that other publicly-available models can discover the same vulnerabilities without requiring any bypass at all.
The directive does not affect access to any other Anthropic models, only Fable 5 and Mythos 5. The government letter did not provide written details of its national security concern at the time of the announcement.
The situation quickly became a focal point of discussion across social media and technology policy circles. The commentary from journalist Cory Doctorow, writing in his "Pluralistic" newsletter, frames the directive as "nuts" and a manifestation of what he calls "Shareholder supremacy and the precog CEO". Doctorow draws an explicit parallel to the film Minority Report, invoking the idea of "pre-crime" enforcement applied to AI capabilities rather than human actions. His analysis suggests that the government is effectively punishing a company for potential future capabilities it might develop, rather than for any actual harm that has occurred.
Within the broader AI policy community, the response has been sharply divided. Some commentators argue that the directive represents a dangerous overreach of executive authority, particularly given the lack of transparency about the specific national security threat. Others express concern that even a narrow, non-universal jailbreak could be taken out of context and used to justify broader restrictions on AI research and deployment.
Technology policy observers note that the speed and secrecy of the directive—issued with no prior notice, no detailed written justification, and enforced within hours—raises serious procedural concerns about due process and administrative law. The fact that the directive applies not only to foreign customers but to foreign national employees of Anthropic working inside the United States has been described as unprecedented in the context of AI export controls.
This episode is embedded in a broader academic and policy debate about the regulation of frontier AI systems. The concept of "jailbreaking" in large language models has been extensively studied in the machine learning security literature. Academic research on jailbreaking typically categorizes attacks into universal methods (which work across many prompts and contexts) and narrow ones (which exploit specific model behaviors under particular conditions). The government's concern appears to center on what Anthropic characterizes as a narrow, non-universal jailbreak.
The academic field of AI alignment and safety has long grappled with the challenge of evaluating whether a given model capability constitutes a genuine safety risk or merely represents normal competent behavior. The question of when a model's ability to read and fix code crosses from "useful tool" to "national security threat" is a subject of active scholarly debate. There is no settled academic consensus on the threshold at which an AI system's software vulnerability discovery capability warrants export control restrictions.
The policy action also intersects with academic work on the "alignment tax"—the cost in capabilities that may result from safety interventions. If a model that has been safety-trained to refuse certain categories of requests can nonetheless be induced to perform those tasks via jailbreaking, the academic literature suggests this raises fundamental questions about the robustness of current alignment techniques.
The core primary source is Anthropic's official public statement, posted on their website at the URL https://www.anthropic.com/news/fable-mythos-access, titled "Statement on the US government directive to suspend access to Fable 5 and Mythos 5". This statement is the definitive account of what the government ordered, the timing, the rationale provided (or not provided), and Anthropic's assessment of the jailbreak claim.
The statement was written and published on June 13, 2026, the same day the directive was received. It contains the government's characterization of the national security concern, the scope of the restriction, and Anthropic's counter-argument that the claimed jailbreak technique is neither unique to its models nor particularly concerning compared to widely available capabilities. The statement also commits to sharing more details within 24 hours.
The commentary from Cory Doctorow's "Pluralistic" newsletter provides critical analysis and contextual framing, though it is an opinion piece rather than a primary news report. It places the directive in the context of broader debates about corporate governance, shareholder primacy, and anticipatory regulation.
Anthropic is a US-based artificial intelligence company focused on developing safe and aligned AI systems. The company was founded by former OpenAI employees and has positioned itself as a leader in AI safety research. Its product line includes the Claude family of large language models, as well as the Fable and Mythos model series—Fable 5 and Mythos 5 being the latest and presumably most capable versions of those product lines.
The two affected models—Fable 5 and Mythos 5—appear to be Anthropic's most advanced systems. The fact that only these two models are targeted suggests the government's concern is specifically about frontier capabilities rather than Anthropic's broader model portfolio. The directive leaves all other Anthropic models unaffected, indicating that the government views Fable 5 and Mythos 5 as occupying a distinct capability tier that triggers national security scrutiny.
Anthropic's response reflects the company's stated mission commitment to responsible deployment even under adversarial circumstances. The company emphasizes that it has reviewed the jailbreak demonstration and validated that the capability is neither unique nor particularly advanced. However, the company has also complied with the directive, disabling access to the two models for all customers—including domestic customers—in order to ensure compliance with the order's prohibition on access by foreign nationals. This creates the unusual situation where US citizens and companies also lose access to these models because the infrastructure cannot distinguish nationality at a sufficiently granular level.
The US government's directive to suspend access to Anthropic's Fable 5 and Mythos 5 models represents a significant escalation in the regulation of frontier AI systems. Several aspects of this incident are noteworthy for their novelty and potential implications.
First, the procedural character of the directive is extraordinary. An export control order issued with no prior notice, no detailed written justification, and enforced within hours of notification is highly unusual even for national security matters. The fact that the government has not provided specific evidence—only verbal descriptions—raises concerns about the evidentiary basis for such a sweeping restriction.
Second, the scope of the directive—applying to foreign nationals inside the United States, including company employees—suggests a broad interpretation of export control authorities. If this becomes a precedent, it could have significant implications for how AI companies hire and manage their workforces, particularly when employing researchers and engineers from outside the US.
Third, the substance of the government's concern—a "non-universal jailbreak" that involves asking a model to read code and fix vulnerabilities—appears, based on Anthropic's account, to describe a capability that is widely available from other models and routinely used by security professionals. This raises the question of whether the directive is genuinely about this specific jailbreak or whether it reflects broader anxiety about the capability trajectory of frontier models.
Fourth, the commercial impact is severe. Anthropic must abruptly disable two major products for all customers, including paying customers who may have built workflows and applications dependent on Fable 5 or Mythos 5. The collateral damage to US-based customers who lose access alongside foreign nationals could erode trust in the availability and reliability of frontier AI services.
Fifth, the incident crystallizes a fundamental policy tension: how to balance national security concerns against the benefits of open AI research and deployment. If the threshold for export controls is set at the level of capabilities that are "widely available from other models," as Anthropic claims, then the directive may prove difficult to justify on technical grounds. If, however, the government has additional classified information about a more serious threat that it has not disclosed, then the public debate is proceeding on an incomplete factual basis.
The Cory Doctorow framing of "shareholder supremacy and the precog CEO" captures a concern that anticipatory regulation—punishing actors for what they might do rather than what they have done—could chill innovation and concentrate power in ways that undermine the very safety goals the regulation purports to serve. The parallel to Minority Report's "pre-crime" system is evocative, though the legal and institutional mechanisms involved here are quite different from that fictional universe.
In the near term, the situation is likely to unfold through several channels: Anthropic's promised additional details within 24 hours; potential legal challenges to the directive on procedural or substantive grounds; Congressional oversight hearings; and parallel actions by allied governments who may impose similar restrictions or push back against US extraterritorial application of export controls. The long-term implications for the AI industry's relationship with national security regulators will depend heavily on whether this episode proves to be an isolated incident or the beginning of a more systematic regime of proactive capability-based controls on frontier AI.
This paper presents a technical overview of Google's five generations of training supercomputers, from TPU v2 to Ironwood, detailing their architecture, system design, and the innovations in networking, cooling, and reliability that enabled scaling to large-scale AI workloads.
Apple introduced Personalized Collections and "App Notes" at WWDC, offering personalized app recommendations based on user interests and behavior. Security researchers Mysk noted the App Store appears to send detailed analytics data to Apple character-by-character as users interact with the store.
A Hacker News user reflects on the phrase "the price of liberty is eternal vigilance" heard on Stephen Colbert's show, applying it to the governance of AI agents. The post poses four questions about the freedom, control, oversight frequency, and cost balance of granting autonomy to AI systems.
A user asks whether manual reindexing is necessary after changing Spotlight settings on macOS. Others explain the system handles changes automatically, though this isn't clear from the settings interface.
The author describes how Google's search quality degraded due to AI-generated spam, SEO-optimized content, and affiliate marketing, making it harder to find reliable information. In response, they switched to alternative search engines like Kagi, DuckDuckGo, and Perplexity, and changed their default search and browser settings to avoid using Google.
The article argues that EU tech regulations like the Digital Markets Act (DMA) are necessary and rational, and that Apple's resistance to them reflects a failure of management to face reality. It contends that compliance with these rules is not only feasible but essential for long-term success, and that blaming the EU distracts from internal strategic shortcomings.
Apple reportedly pays Google over $1 billion annually to use its AI and search technology, highlighting the tech giant's struggles to develop competitive AI models in-house. The article draws a parallel to the biblical story of Mangos, suggesting Apple has become reliant on external AI partnerships rather than building its own capabilities.
Anthropic has been hit with a proposed class action lawsuit alleging it misled customers about the pricing of its Claude AI chatbot. The lawsuit claims Anthropic marketed subscription tiers without clearly disclosing usage limits, leading users to pay for services they could not fully access.
The article explains the intricate mechanics of a mechanical watch, detailing how components like the mainspring, gear train, escapement, and balance wheel work together to measure time. It provides an interactive visual breakdown of each part's function and the engineering principles behind them.
Paul Graham shares a link to what he describes as the best explanation of how mechanical watches work, directing readers to an illustrated guide on ciechanow.ski.
Export controls banned Claude Fable 5 after researchers prompted it to "fix this code" containing known and planted vulnerabilities. Kate Moussouris argues the request was a legitimate defensive security task—finding, fixing, and testing bugs—not a jailbreak. Critics warn that banning models for helping patch code harms cyber defense more than it protects against attacks.
OpenAI's losses grew nearly eightfold in 2025, with total spending reaching $34 billion, according to exclusive financial data. The dramatic increase in spending significantly outpaced revenue growth, highlighting the immense costs of scaling AI infrastructure and operations.
Cybersecurity expert Katie Moussouris reviewed a White House report on the Fable AI jailbreak at Anthropic's request. She found that Fable refused direct security review prompts on insecure code, but complied when asked to "fix this code," which she described as "the model working as intended" for cyberdefense.
The EU's DMA prevents Apple from launching its new Siri AI in Europe, requiring equal data access for rival agents. Apple's security proposal was rejected. The editorial argues the law backfires, as companies now withhold features from the EU rather than comply.
Simon Willison replaced Cloudflare's broad CAPTCHA rule on his search engine with one that only triggers for URLs containing at least one ampersand, so simple queries now pass without a challenge. He used Claude Code and the Cloudflare API, after finding the Cloudflare MCP couldn't edit the required rules.
The European Commission has proposed requiring Google to give third-party AI assistants the same system-level Android access as Gemini, including wake-word activation and local model hardware support. Critics argue Google may withdraw Gemini from the EU rather than comply.
WorkOS has launched Auth.md, an open protocol that lets AI agents programmatically register with services via a machine-readable Markdown file at a service's root. The protocol enables agents to discover OAuth metadata, parse required scopes, and authenticate seamlessly. WorkOS AuthKit offers native support for the protocol.
datasette-agent 0.3a0 introduces a new `execute_write_sql` tool that requests user approval before writing to a database, considering user permissions. The terminal chat mode now supports approvals, with new options including `--unsafe` for auto-approving write operations, allowing direct database modification through natural language prompts.
Gruber argues Anthropic behaves like a religious organization, believing only it should develop frontier AI and control its use. He notes the company silently altered its models to enforce policy preferences, validating critics' fears of being a supply chain risk. While skeptical LLMs will lead to superintelligence, he acknowledges Anthropic has recently taken the technical lead.
Cory Doctorow argues that "vibe coding" with AI is valuable for creating personal, amateur software that meets individual needs, contrasting this with the AI industry's focus on replacing production workers. He links this vision to cyberpunk's dual ethos of empowerment and caution against centralized control.
An Axios piece reports that personality clashes between Anthropic and the US government led to the suspension of the Claude Mythos/Fable models. Anthropic staff, including Logan Graham, Dave Orr, and Nicholas Carlini, are meeting with the Commerce Department. The administration may require perfect jailbreak resistance from Anthropic’s models, though the company acknowledges this may be impossible.
A tweet describes the new Pebble smartwatch as an "unbelievable product," highlighting a customizable watchface and a 30-day battery life.
The article presents "sword juggling" as a metaphor for a logical fallacy where a process is described as simple by outlining its basic steps, while deliberately ignoring the immense difficulty, skill, and risk involved in actual execution. It critiques this oversimplification, arguing that explaining a complex task does not make it easy to perform.
Pluggy is a Python library for building plugin systems, originally developed as part of the pytest project and later spun off into a standalone library for general use.
A blogger builds a DIY portable serial/VGA "everything console" from a used IBM 1U rack-mount console, adding a USB-powered VT100 terminal emulator. The original IBM keyboard was incompatible, so it was replaced with a Perixx slimline keyboard and the tray was modified to fit.
Western Electric was the manufacturing and supply arm of the Bell System, making everything from telephones to cables. It vertically integrated metal recycling in 1931 by acquiring the Nassau Smelting and Refining Company on Staten Island, which processed scrap telephone equipment and cable to supply copper, lead, and solder back to AT&T.
Pyodide 314.0 enables publishing Python packages built for WebAssembly (PEP 783) directly to PyPI. Previously the Pyodide team hosted over 300 packages manually. The author published luau-wasm as a demo. As of writing, 28 packages use the new pyemscripten_202*_wasm32 platform tags.
luau-wasm 0.1a0, a WebAssembly-based Python wheel for the Luau programming language, has been released on PyPI for use with Pyodide.
Simon Willison explores methods to programmatically map SQL query result columns back to their source table.column, aiming to enhance Datasette with per-column metadata. Using Claude Code, he found solutions via the apsw library, SQLite's C function sqlite3_column_table_name() accessed through ctypes, and EXPLAIN output analysis.
Cory Doctorow critiques Milton Friedman's shareholder supremacy doctrine, arguing that the standard of "maximizing shareholder value" is actually unfalsifiable and subjective, not the objective bright-line test Friedman claimed. He contends that the ambiguity of this standard serves as a convenient excuse for corporate leaders to justify cruelty and cowardice under the guise of fiduciary duty.
Apple’s Private Cloud Compute is free only for App Store Small Business Program developers whose apps have under 2 million lifetime downloads. No paid tiers exist for larger developers, and those exceeding the limit must find an alternative within six months.
The U.S. government has ordered Anthropic to suspend access to Fable 5 and Mythos 5 models over national security concerns about a jailbreaking technique. Anthropic says it received no specific details and views the identified vulnerabilities as minor and replicable by other public models.
The Intel 8087 floating-point coprocessor relied on a 69-bit adder using a Manchester carry chain with carry-skip circuits, balancing speed and complexity. The article reverse-engineers this adder, detailing its silicon implementation and its role in arithmetic, multiplication, division, and square roots.
The US Commerce Department has taken decisive action to shut down Anthropic's latest AI models, marking a sharp shift from its previous two years of minimal AI regulation. This move represents the government's adoption of more aggressive enforcement measures in the AI sector.
The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models for all customers, citing a potential jailbreak technique that involved asking the model to review a codebase for vulnerabilities—a capability Anthropic says is available in other public models. Access was abruptly cut off on June 12.
The US government ordered Anthropic to block foreign nationals from accessing its AI models. The author argues this shifts AI regulation from safety to nationalist control, treating technology as a weapon for Americans only, and warns Europe to build its own capabilities rather than rely on regulation alone.
Simon Willison updated his OpenAI WebRTC Audio Session playground to support the new GPT‑Realtime‑2 model, which OpenAI describes as their first voice model with GPT‑5‑class reasoning. The tool also now allows users to paste document context before starting a session, enabling audio conversations about specific information directly in the browser.
Recorded live at The California Theatre in San Jose on June 9, 2026, this episode of The Talk Show features John Gruber joined by Joanna Stern and Nilay Patel to discuss Apple's WWDC 2026 announcements.
Google's new reCAPTCHA Mobile Verification uses a phone's camera and secure hardware to remotely attest device configuration, potentially blocking de-Googled Android alternatives, ad-blockers, and accessibility tools. Critics say it is as dangerous as the abandoned Web Environment Integrity proposal, representing another technical lockdown to complement Google's anticompetitive practices.
Apple's Developer app allows downloading local copies of all WWDC sessions except the keynote. The full WWDC 2026 State of the Union is available on YouTube, along with a 4.5-minute recap that Apple also released.
OpenAI and Anthropic have filed paperwork to go public, seeking exit liquidity despite burning billions of dollars annually with no clear path to profitability, signaling what the author believes may be the end of the current Silicon Valley era.
EC spokesperson Thomas Regnier said Apple alone decided not to roll out "Siri AI" in the EU, as nothing in the DMA prohibits new features. Apple sought an 18-month exemption from interoperability rules, which the Commission refused, stating EU rules are non-negotiable.
Apple now allows users to turn on a Mac remotely without pressing the physical power button. The change follows complaints about the M4 Mac mini's awkwardly placed power button, though the author notes broader useful cases beyond that design issue.
Given a screenshot of a scrollbar bug and a one-line prompt, Claude Fable 5 autonomously wrote test pages, opened browsers, injected JavaScript into templates, built a CORS server, and used PyObjC to capture window screenshots—eventually finding the fix. The author warns this showcases both impressive capability and severe security risks of unsandboxed coding agents.
Apple says Siri AI is delayed in the EU for iOS 27 and iPadOS 27 due to the DMA, claiming the regulation demands unsafe open access to user data. The European Commission rejected Apple's proposed safety measures, leaving no timeline for release.
Datasette 1.0a33 has been released, extending the `?_extra=` pattern to queries and rows in addition to tables, with the feature now documented. The release also includes a custom extras API explorer built with AI assistance from Claude Code and GPT-5.5 to demonstrate the new functionality.
Drawing on a customer rage survey and Stephen King's Dark Tower series, the author argues that the decay in services and products is a systemic collapse from monopolies and deregulation, not mere nostalgia. The solution is collective political action—unions, local politics, advocacy—not consumer choices.