Anthropic Fable 5: a Marketing Platform Gone Wrong

Has Anthropic poisoned its own well by arguing its new Mythos model was so dangerous it couldn’t be released publicly?

After Amazon researchers discovered a jailbreak of Fable 5 (Mythos with guardrails), the U.S. government banned use of both Mythos and Fable 5, exactly the kind of government scrutiny that could disrupt Anthropic’s  business most.

Amazon CEO Andy Jassy was reportedly the source of security concerns that led the U.S. government to force Anthropic to shut down Fable 5 and Mythos 5 for every customer, also imposing an export control ban on both models.

Oddly enough, Amazon is one of Anthropic’s largest investors, having put in billions and receiving a $100 billion cloud spending commitment in return. 

Anthropic says the jailbreak technique surfaced “a small number of previously known, minor vulnerabilities.” 

It called the government’s response disproportionate and said the capabilities causing concern are already available in other publicly accessible models. The shutdown affected every customer globally because Anthropic cannot filter foreign nationals from U.S. users in real time.

For Anthropic, the immediate question is how quickly it can restore access. And that will happen.

For the broader industry, the precedent is what matters. Anthropic has been lobbying for more regulatory power over AI. It just found out it might get what it wants. 

Anthropic's own CEO publicly called for the government to have authority to block dangerous model releases. The government then used such calls against Anthropic two days later.

Optimists might argue that government restrictions on dual-use technologies might slow down technology diffusion, but cannot halt it.

The general pattern across technology history is this: governments impose controls with genuine security rationale, those controls prove partially effective in the short term, generate substantial unintended costs, and are eventually relaxed or circumvented.

The regulated party frequently bears costs that accrue to unregulated competitors.

Technology	Regulation	Period	Short-Term Effect	Long-Term Outcome
Cryptography, Encryption (PGP, SSL)	U.S. export ban; treated as a munition under ITAR; Clipper Chip mandate attempt	1975 to 2000	Phil Zimmermann prosecuted 3 years for posting PGP online; RSA export restricted New America · Brookings	Controls collapsed when it became obvious software could not be bottled; Clinton-era deregulation followed. In subsequent decades, virtually all predictions about encryption's economic benefits were borne out — SSL, SSH, electronic banking, e-filing, and VPNs depended on exactly the strong encryption the government tried to restrict.
Cryptography	Export controls treated software as equivalent to munitions	1990s	The Zimmermann case dragged on three years before being dropped, and export control laws were eventually rewritten after it became obvious that software couldn't be contained like rocket motors. Crowe LLP · Wassenaar/arxiv	The EU abolished cryptographic export controls within the union in 2000, a decision subsequently adopted by the Clinton administration; more recent trends have involved further relaxation rather than tighter controls.
CoCom, Cold War Computing	Coordinating Committee for Multilateral Export Controls: Western embargo on technology exports to Soviet Bloc	1949–1994	In the period 1951–1967 CoCom performed reasonably well despite its known limits. After that, enforcement eroded. ChinaTalk/Ottinger · Texas Nat'l Security Review	CoCom's effectiveness was reduced throughout its history by overt non-compliance, differences between member nations, the secretive regime, and the financial incentives of tech exporters. CoCom did not prevent the USSR from accessing key technologies; the size of the control regime expanded over time, weakening enforcement and encouraging defection. Disbanded 1994.
Toshiba-Kongsberg Scandal	Sanctions on Toshiba for selling machine tools enabling quieter Soviet submarine propellers	1982–1987	Washington imposed sanctions on Toshiba worth $30 billion amid congressional claims the breach jeopardized U.S. national security. AFSA	The $30 billion estimate was primarily based on a hyperbolic claim that the entire U.S. nuclear submarine fleet would need to be replaced — the actual magnitude of the damage remained unclear. Sanctions were eventually relaxed; Toshiba survived.
Nuclear Technology	Nuclear Non-Proliferation Treaty (NPT); Nuclear Suppliers Group; export controls on enrichment technology	1968–present	By the 1950s, it became clear that pure denial had neither stopped the Soviet Union nor the UK from acquiring nuclear weapons. The 1954 revision of the Atomic Energy Act reflected a shift from prevention-through-denial to influence-through-cooperation. CNAS · AADS	U.S. dominance in commercial nuclear energy in the 1970s allowed the U.S. to keep Taiwan and South Korea from developing their own nuclear facilities — a genuine success. But Pakistan, India, North Korea, and Israel all acquired nuclear capability despite controls. NPT remains the most durable tech-control regime, though partial.
Semiconductors, AI Chips (China)	BIS October 2022 controls on NVIDIA A100/H100 and manufacturing equipment; tightened 2023, 2024, 2025	2022–present	Implementation significantly disrupted China's semiconductor ecosystem, causing price spikes for some device types and forcing workforce reductions. CSIS/HSToday · CSIS	Loopholes, alternative approaches, and unforeseen outcomes have diminished long-term efficacy — the very restrictions designed to hinder Beijing may instead be accelerating China's domestic progress. Companies impacted by the October 2022 controls, on average, even outperformed comparable unaffected companies, exhibiting higher increases in R&D spending and patent filings — a counterintuitive outcome driven by AI chip demand.
Semiconductor Controls	EAR controls tightened repeatedly; allied coordination with Netherlands, Japan	2022–present	According to research, versions of Intel's Xeon Gold and NVIDIA GeForce RTX chips can be bought on Taobao; students at Tsinghua report being able to "easily circumvent" restrictions on U.S. EDA software. Texas NSR	CoCom did not prevent the USSR from accessing key technologies; the current regime is similarly porous, and China is a more adept target. Long-term verdict still open.
AI Models Fable 5 / Mythos 5	U.S. Commerce Dept. emergency export control directive; all foreign nationals barred	June 2026	Anthropic had to cut off access to both models for all customers worldwide because it could not rapidly implement nationality-based filtering. Anthropic disputed the severity of the jailbreak, arguing many of the same vulnerabilities could be discovered using other publicly available models. Fortune · Axios · State of Surveillance	Too recent to assess long-term outcome. The next letter may go to a different vendor; the policy tools used here will be available to every future administration. Watch for restored access accompanied by new conditions such as nationality verification, tighter pre-deployment review, or federal observer access.

Several durable findings emerge from this body of evidence:

• Unilateral controls erode without multilateral coordination

• The control regime tends to expand and become self-defeating

• Denial often accelerates indigenous capability in the target

• Software and information resist containment

• The regulated party bears costs; competitors do not. 

Rival Sam Altman, of OpenAI, says Anthropic’s fear-based marketing now has come back to bite Anthropic. 

AI Model Pricing Eventually Moves to a Cloud Model

So far, language model pricing based on usage closely mirrors the early patterns of cloud computing, .

Which matured into a massive, lower-margin commodity market with heavy optimization practices.

AI inference is on a similar but possibly accelerated trajectory.

We might reasonably expect:

• Price deflation and commoditization

• More sophisticated pricing models

• Shift to agentic operations

• Hybrid platforms

• More cost governance measures.

source: Rohan Modi 

It is reasonable to expect prices for baseline capabilities will fall (potentially another 5-10x over years).

Frontier capabilities will remain premium but get cheaper over time, as well.

We probably should see "good enough" models for most tasks, mirroring how basic cloud virtual machines became very affordable.

Reserved-like plans, volume discounts and sustained use discounts already are emerging, with more pricing granularity. 

The shift from chatbots to agents will likely lead to pricing based on:

• Per-agent/subscription (like "digital employee" salaries)

• Outcomes (pay per resolved ticket, qualified lead, successful workflow)

• Hybrid models (base + usage + performance bonuses).

Overall, the shift is from pricing based on “raw compute” to “managed services” with value-based elements.

source: Rohan Modi 

Overall, expect AI pricing to become more predictable, value-aligned, and cheaper per capability unit, much as cloud computing evolved. 

SpaceX has Gone Public

SpaceX has completed the largest initial public offering in history, raising $75 billion. The listing priced 555.6 million shares at $135 each, granting the company a mammoth market valuation of $1.77 trillion.

source: Bret Jensen 

Source: SpaceX, the Verge 

source: SpaceX, Space News 

And even if Starlink powers today’s revenue, the future upside is clearly pegged to artificial intelligence, according to SpaceX itself. 

Will the 2026 World Cup Create Any Long-Term Economic Benefit for Host Nations?

World Cup long-term economic effects will be negligible, economists at Goldman Sachs say. That might seem unlikely, given the 2026 FIFA World Cup featuring 48 teams and 104 matches across the United States, Canada and Mexico.

After Goldman Sachs International economists Kevin Daly and Mambuna Njie studied gross domestic product data covering every World Cup since 1982, they find hosting produces a marginally positive but statistically insignificant effect on real output, with long-run impact that is effectively zero.

FIFA and the World Trade Organization disagree.  A joint study they published in April 2025, developed by consultancy OpenEconomics, projects a $17.2 billion contribution to U.S. GDP, $30.5 billion in gross output and approximately 185,000 full-time equivalent jobs for the host country alone. 

Across all three host countries, the combined GDP estimate reaches $40.9 billion, the report argues. 

Different methodologies help explain the differences. 

Beer, merchandise and apparel purchased in their own markets does not register in U.S., Canadian or Mexican GDP. 

Domestic spending on World Cup-related goods and services may simply be redirected from other consumption categories rather than representing new activity.

There is short-term lift, but no lasting contribution.

Leakage effects also are real: profits from international licensing, sponsorship and supply chains accrue outside the host country’s GDP.

On the other hand, it stands to reason that several industries should benefit, including:

• European and US consumer staples (brewing companies including AB InBev, Molson Coors, Constellation Brands, Heineken and Carlsberg)

• European consumer discretionary, primarily sportswear (the ones we know: adidas, PUMA)

• U.S. retail and softlines (Academy Sports + Outdoors, Dick’s Sporting Goods, Nike)

• U.S. lodging and leisure (Hyatt, Marriott, Hilton, Airbnb)

• U.S. airlines. 

There will be significant industrial impact in those segments of the market, to be sure. Concentrated, time limited but real. 

Wider and long-term benefits will likely be negligible, if measurable at all. 

In many ways, the impact is similar to that supposedly created by municipally-financed sports stadia. 

The claim that government-financed sports stadia act as engines for economic growth is widely contested within the field of economics. 

While proponents often cite job creation, increased tax revenues, and regional prestige as primary justifications for public subsidies, empirical research consistently demonstrates that these facilities rarely produce significant, measurable net economic benefits for host cities.

The core economic argument against public financing centers on the substitution effect. 

Economic models often fail to account for the fact that a large portion of spending at a stadium is not "new" money introduced into the local economy; rather, it is money that residents would have otherwise spent on other local entertainment options, such as restaurants, movie theaters, or other cultural activities. 

Because this spending is simply redirected, there is little to no net increase in total local economic activity.

Furthermore, economic impact studies commissioned by proponents often rely on flawed multipliers that exaggerate the stimulative effect of sports expenditures. 

These studies frequently ignore "leakage," where significant portions of the revenue (such as players' salaries) are exported out of the local economy because athletes and owners often do not reside in the city where they play. 

Consequently, most independent academic research concludes that the public cost of these subsidies far exceeds any marginal economic growth they may stimulate.

Study / Authors	Findings	Source
Bradbury, Coates, & Humphreys (2023)	Retrospective analysis confirming that stadiums are poor public investments and that public outlays provide meager benefits.	Link
Matheson (2018)	Found no evidence that stadium subsidies yield economic growth; suggested that at most 5–15% of public cost might be justified by "public good" (civic pride).	Link
Bradbury (2022)	Found negligible net increases in sales tax collections and noted that approximately one-third of stadium sales displace other local activity.	Link
Siegfried & Zimbalist (2002)	Demonstrated that standard impact multipliers exaggerate benefits by over 400% due to consumer substitution and economic leakage.	Link
Coates & Humphreys (2003)	Econometric analysis finding no evidence of positive economic benefits associated with stadium construction; some results indicated negative impacts.	Link

U.S. Productivity is Rising, but AI Doesn't Seem the Reason

U.S. productivity has been rising for several years, but artificial intelligence is probably not the reason, at least, not yet. 

According to a report published by the Federal Reserve Bank of San Francisco, the U.S. economy expanded at a relatively steady pace of around 2.5 percent per year over the past three years, even though employment growth slowed to near zero. 

Almost by definition, higher output with the same input means higher productivity. But it is not clear artificial intelligence has much to do with the increases.

A survey of nearly 6,000 senior business executives in the United States, United Kingdom, Germany and Australia published by the National Bureau of Economic Research found:

• 69 percent of firms actively use AI

• 66 percent of executives regularly use AI

• Average use is about 1.5 hours a week

• 90 percent of executives report little own-firm impact of AI over the last three years

• 90 percent report no impact on employment or productivity

• Over the next three years, respondents predict that AI will boost productivity at their firms by an average of 1.4 percent

• Will raise output 0.8 percent

• Cut employment 0.7 percent

• Employees believe AI will raise employment 0.5 percent in the next three years.

Perhaps the most-unexpected result is the employee belief that AI will actually boost employment at their firms over a three-year period. That findings seems at odds with the usual press reports suggesting employee angst about AI impact on employment. 

The least-surprising result should be the inability to pinpoint AI productivity gains. 

For starters, U.S. productivity has recently been rising since about 2019, well before AI emerged as a potential driver. 

Labor productivity measures how efficiently workers use the capital available to them, such as equipment or software. The data suggests workers are doing so. 

Total factor productivity uses a broader view, measuring how efficiently the economy uses all inputs together, including both labor and capital.

One interpretation might be that workers have been using tools effectively, but that the gains have not yet shown up in TFP metrics. 

Think about your own work. Many of us would absolutely agree that AI has boosted our own personal productivity. But few of us can point to measurable gains in economic `outputs. 

Federal Reserve Bank 

And U.S. productivity had been rising since about 1992 as well, to 2000. 

Federal Reserve Bank 

For some observers, past experience suggests a productivity gain will happen. The U.S. economy has experienced several distinct productivity regimes over the past 70 years, including a high-growth period in the late 1990s, with the proliferation of computers and the internet, and a lengthy period of low average growth during the 2010s.

Federal Reserve Bank

Right now, it appears there is a significant disconnect between labor productivity and TFP. 

“The divergence between strong labor productivity growth and more modest TFP growth suggests that recent investments related to AI might be making workers more productive by providing them with better tools, such as new software and expanded computing capacity, but broader efficiency gains remain unrealized so far,” the report says.

But the report also says the pattern (Labor productivity and TFP misaligned) resembles what we saw when the internet became important. 

There was a lag then, and there is arguably a lag now. As the adage goes, one can see the impact everywhere but in the outcomes (paraphrasing the Solow Paradox: "You can see the computer age everywhere but in the productivity statistics.").      

Orbital AI Compute Seems to be Coming, but Not at Scale, Right Away

With SpaceX going public on June 12, 2026, lots of investors will be pondering the feasibility of creating orbital data centers at scale.

But space-based data centers are not an immediate replacement for terrestrial data center alternatives for reasons of initial cost and capacity. Launch costs remain substantial.  

Potential upsides center on lower ongoing costs offsetting high upfront costs, eventually, though initial total operating costs will probably not match terrestrial alternatives:

• Cheap/abundant power: Solar in orbit provides ~36% higher irradiance, near-constant supply (no night/clouds/weather), and very low marginal costs (projections ~$0.005/kWh vs. $0.04-0.08/kWh terrestrial wholesale). No grid connection, fuel, or large storage needed in ideal orbits.

• Lower OpEx: Projections include 97% lower operating costs in some models (energy + cooling). No land, permitting, property taxes, or water for cooling. Avoids terrestrial delays/queues for power infrastructure.

• Scalability and utilization: Unlimited "land" in orbit for expansion. High utilization from constant power. Falling launch costs could lead to cost parity or better for power-dominant workloads by late 2020s to 2030s.

Orbital systems could ease some important terrestrial obstacles:

• Energy and emissions: Relies on clean solar (potentially 10x lower CO₂ emissions). Reduces strain on terrestrial grids, which often use fossil backups for data centers.

• Resource Savings: No water consumption for evaporative cooling (a major terrestrial issue). Frees land for other uses; avoids local ecosystem/power price impacts from hyperscale farms.

• Overall footprint: Could lower terrestrial data center growth, helping with power queues, water scarcity, and NIMBY opposition.

Of course, environmental impact is still there. Launch emissions, space debris (cluttering orbits, potential Kessler syndrome risk), manufacturing impacts and end-of-life disposal remain issues. 

Some use cases might make more sense. Workloads tolerant of moderate latency (~100-500 ms round-trip) and benefiting from proximity to space data or constant power suggest suitability for:

• AI Inference: Querying trained models (chat, search, voice agents, video generation, back-office automation)

• Some telemetry use cases: Onboard near-source analysis of Earth observation, climate monitoring, disaster detection (wildfires/floods), maritime surveillance, sensor apps

• Some edge compute cases: Real-time processing for satellites, space cybersecurity or autonomous operations or resilience against terrestrial outages/disasters

• High-Security/ Sovereign Compute: Isolated environments for sensitive data, national security, or regions with poor terrestrial infrastructure.