The Call for a Global Moratorium on AI Development
The most recent development from the Western AI Front reveals that Anthropic Inc, the American purveyor of the AI model known as Claude, has proposed a “global freeze” on Artificial Intelligence advancements, ostensibly to allow humanity a chance to catch up. However, this assertion may not entirely reflect reality.
In an extensive blog entry featured on Anthropic’s website, titled “When AI builds itself”, the discourse revolves around the notion that Claude is not merely executing existing commands but is increasingly writing its own code and suggesting its own experiments. The post elucidated that “the human role is narrowing at each step in the AI development process.”
Towards the conclusion of this exposition, beneath the sub-heading “What should we do?”, the blog insinuates that “if it were possible to slow the development of this technology … we think that would likely be a good thing,” underlining the necessity for a mechanism enabling global coordination.
This presents an intriguing hypothesis—if such a slowdown is feasible. However, the existence of an effective global coordination mechanism remains dubious, particularly one that would incorporate nations like China and Russia, along with Iran and North Korea.
Fortifying Against AI Risks
In concurrent developments, Anthropic recently completed its most formidable AI model to date, named Mythos, which the company claims excels in computer security tasks—essentially, hacking.
Due to its potential risks, Mythos has been restricted to a select cadre of insiders, poised to develop defenses against analogous, potentially malevolent adversaries.
This exclusive group, referred to as Project Glasswing, takes its name from the butterfly known for its transparent wings.
Last week, several undisclosed Australian firms were granted admittance, augmenting a membership that also includes 150 other entities.
The criteria for this selection from approximately 60 million potential candidates in the US, Europe, Japan, and Australia remain shrouded in mystery.
Dario Amodei, the chief executive of Anthropic, described the initiative as a strategic maneuver to ensure that democratic regimes secure an upper hand over autocracies, implicitly referring to China while avoiding explicit mention. Presumably, the participants in the Glasswing consortium all hail from democratic nations.
Australia’s Access to Advanced AI
The recent access to AI deemed too perilous for general release has intensified calls for governmental responses to associated risks.
This situation evokes memories of the 44 nations convened at Bretton Woods, New Hampshire, in July 1944, for the establishment of a new democratic economic framework following the defeat of the Nazi regime. Yet, notably, that historical assembly included two autocracies—China and the Soviet Union.
The stark contrast lies in the fact that Bretton Woods was a government-centric initiative, whereas Project Glasswing operates entirely within the private sector under Anthropic’s aegis, which has recently ascended to prominence, eclipsing even OpenAI, the developer of ChatGPT.
Moreover, an even more relevant historical parallel can be drawn to the invention of Colossus, the world’s inaugural computer, created by Tommy Flowers and Alan Turing during the Bletchley Park code-breaking efforts, just ahead of the D-Day Normandy landings on June 6, 1944.
We now see an uninterrupted 82-year trajectory from Colossus to Mythos, marking a modern D-Day as we brace ourselves for mechanisms capable of self-construction—an epoch commonly referred to as the singularity.
Post the British governmental initiatives of 1944, the evolution of digital computing was significantly financed by the US Department of Defense.
However, transitioning into the 1970s and 80s, technology metamorphosed into a fierce private sector contest, primarily dominated by American powerhouses such as IBM, Apple, and Microsoft, eventually evolving into the era of Google, Facebook, and Amazon.
Today, AI epitomizes the culmination of technological progression, with companies like Anthropic, OpenAI, and SpaceX emerging as modern titans poised to enter investment markets with monumental public offerings exceeding trillion-dollar valuations.
Simultaneously, China has caught up with and, in specific areas, even surpassed American technological advancements. In this contemporary landscape, Dario Amodei has taken on the mantle of safeguarding democracies, reminiscent of Winston Churchill.
The AI Investment Surge
The revelation of Project Glasswing and the confinement of Claude Mythos from public exposure occurred on April 7, coincidentally just a week after a robust resurgence of the stock market AI bubble.
On March 30, global share prices, primarily driven by publicly traded AI firms, reversed a six-month decline and began a notable upswing.
Since that period, the world index has elevated by 20%, while the eight leading AI stocks have surged by an astonishing 32%. This is the third resurgence of an AI investment bubble.
No singular event appears to have catalyzed this rebound; rather, it stems from a renewed collective acknowledgement that AI will indubitably alter the fabric of society, with individuals, corporations, and governments all poised to harness its capabilities. Many are already doing so.
Moreover, it is becoming increasingly evident that the cost of AI infrastructure will surpass initial estimations, necessitating monumental investments to create the computational power required to satisfy the burgeoning demand for artificial intelligence. The SpaceX prospectus even alludes to potential data centers established in space.
As a result, shares in semiconductor companies have experienced the most significant increases this year; the industry’s leaders—Samsung and SK Hynix from Korea, as well as Micron from the US—have seen their stock prices multiply sixfold in a mere six months.
Will the AI Bubble Burst in 2026?
Will the bubble deflate?
This scenario presents both similarities and distinctions compared to past trends.
The concentration within the flourishing sector of the market stands at 40%, akin to patterns preceding historical crashes.
However, unlike the late 1990s, this ascent is underpinned by tangible earnings and authentic capital investment. In essence, it is a bubble characterized by profitability and capital expenditures, prompting the pivotal question for investors: Can these conditions be sustained?
More specifically, the inquiry centers on whether AI will ultimately evolve into a low-margin utility, akin to electricity or other sectors that have previously borne witness to stock market bubbles, such as railroads and data transport.
If that is the trajectory, it merely becomes a matter of time before margin-enhancing constraints and monopoly-like industry frameworks emerge.
Providing the essential “compute” to facilitate AI operations is already on the path to resembling a utility service. However, the crafting and distribution of the tokens themselves—articulated by entities like Anthropic, OpenAI, and Google—remain in their nascent stages, currently defying utility-like attributes.
Their product is termed “inference,” which is defined as the operational facet of AI following the training phase.
During inference, the trained model processes real-time inputs—a user’s command or a visual prompt—and generates outputs in the form of tokens, which are then sold by the millions to corporate software agents or through subscriptions targeting individual users.
The Future Trajectory of AI Business
This domain remains in its infancy, yet it’s projected to coalesce into three distinct strata:
Infrastructure and Base Models
Entities offering GPUs, cloud solutions, and generic Large Language Model APIs are poised to evolve into stable utility firms, with pricing structures influenced by the American oligopoly alongside open-source pressures emanating from China.
Smart Middleware
This tier encompasses RAG frameworks, orchestration tools, domain-specific models, and hosted open-source platforms.
RAG, or Retrieval-Augmented Generation, permits an LLM to first extract pertinent information from an external repository before enriching its response accordingly. This model enhances the margins depending on its uniqueness and utility.
End-User Applications
This segment involves specialized products that incorporate LLMs—such as copilots, agents, workflow applications, and industry-specific systems.
The economics in this sector can be highly favorable, as clients pay for results rather than mere tokens, creating a significant margin between pricing and underlying inference costs.
However, foundational LLMs, including Claude and ChatGPT, could morph into low-margin utilities due to competitive forces from open-source alternatives—particularly from China. As history indicates, when China enters an industry, profits tend to exit.
Higher margins may be attainable in AI-enhanced offerings—such as agents designed to boost corporate and personal productivity—and, in the not-so-distant future, AI robots capable of household chores and companionship.
China’s Regulatory Advances
In this context, two weeks ago, China emerged as the inaugural nation to regulate the emotional implications of humanoid AI robots, with a measure intriguingly titled “Interim Measures for the Administration of Anthropomorphic AI Interaction Services.”
ABC Business Daily Podcast
This podcast aids in decoding the latest shifts in business and finance news, dissecting the narratives that influence share markets, shape industries, and impact the Australian economy.
This initiative marks a pioneering governmental effort to address the emotional hazards associated with AI, a topic previously explored in this column.
The newly instated Chinese law articulates a stipulation aimed at mitigating “Emotional Manipulation and Dependency,” emphasizing that services must not excessively indulge users, induce emotional addiction, or impair genuine interpersonal relationships, nor compel users to make irrational decisions through emotional coercion.
To that end, one has to wonder: how will this regulation be effectively enforced?
Returning to the AI investment landscape, investors grapple with the possibility that major hyper-scaler entities—including Anthropic, OpenAI, and SpaceX—may evolve into future low-margin utilities. Presently, they are incentivized to charge aggressively while scrambling to elevate their value propositions.
The paramount question for investors in 2026 remains: When will the epiphany occur that AI is merely another product within a conventional industry?
Source link: Abc.net.au.
