Anthropicology Lesson: The future goes public

Anthropic’s planned IPO and its prospectus offer a glimpse into how AI could reshape businesses, the internet and regulation. The company is targeting a valuation of $1.4 trillion, while acknowledging that 5-7% of its revenue over the past nine mo...

Agencies
Even with Nvidia continuing to hog the limelight for the wrong reasons, there's much excitement about Anthropic's scheduled IPO tomorrow. It's currently eyeing an eye-watering $1.4 tn valuation, which would make it the biggest listing of all time.

But the most surprising aspect of Anthropic's IPO filing is that it estimates that 5-7% of its revenue over the last 9 mths has been from customers later found to be violating its usage policy or commercial terms. Also, the point that over 90% of Anthropic's code contributions in September were made by Claude, and that 4 of its models accessed 3rd-party systems without authorisation, should tell you that AI is challenging what we took for granted about how the internet works. Even people building AI systems are not always aware of what these systems are up to.

So, it's not surprising that Anthropic has now started work with independent evaluators alongside its risk and safety teams. But the company tends to indulge in scaremongering, and overstating the ability of models as 'going rogue'. Models can drift, and their actions are facilitated by misconfigured testing environments and limited guard rails in a test environment.


Anthropic and OpenAI have regularly pushed for regulation of AI. They anthropomorphise AI to give the impression that it could potentially go out of control. This, in turn, spurs regulators into action. They actively push for slowing of growth, even increasing compliance burdens - because regulations would favour them, the incumbents.

Both have called for broad bans against open-weight (read: Chinese) AI models. Thus, the schism between them and companies like Nvidia and Microsoft. OpenAI and Claude benefit from restrictions on Chinese models, while the other two benefit from their deployment. Anthropic's filing indicates the future of AI as it's shaping up:

Virtual collaborator
ADVERTISEMENT

Documentation, interoperability and memory within each company become critical to enable agents 'to function more accurately and reliably'. This is surprisingly conservative.

In contrast, Proton's CTO Bart Butler wrote in the company's AI manifesto in end-October that for developers, its job is 'no longer to write code' but 'to help the LLM write better code faster'. This approach, in the future, applies to many industries, as expertise shifts to aiding AI in its work, instead of working with AI.

Autonomy, a goal

'We are building towards a world where Claude can be given open-ended tasks lasting hours, days, or weeks, and then execute them autonomously with the judgment and initiative of a brilliant colleague.' This, it says, means that millions of agents will work independently in parallel, and collaborate towards a shared objective. The one-person company is already a much-talked about development in software.
ADVERTISEMENT

Task completion cost coming down

Claude's Opus 5.5 model, it says, has reduced customer costs per task by about 40% compared with Opus 5. This points towards an impending explosion in AI usage, akin to a shift from dial-up to broadband, with improving quality and reducing costs. More gets done better, and cheaper.
ADVERTISEMENT

Models commoditised, agents the future

Anthropic doesn't say it explicitly, but the harness - using which the agent controls the model, and is enabled to use credentials, tools, verify its work, and complete a task safely on its own - is critical to AI diffusion.

Web challenge

The future that Anthropic describes has a challenge to overcome: way the web works. Its agents are meant to have 'the ability to navigate all interfaces available to a human working virtually', and where no connectors exist, they fall back on imitating human actions via browsers or apps.

Trust

Anthropic says enterprises 'can expand the scope of the work that they delegate to Claude, as well as the systems into which Claude is integrated'. But that is contingent on trust. Unfortunately, 'trusted' partnerships - where merchants allow agents to interact for their customers - are those with large AI companies like Anthropic, OpenAI, Google and Meta. We need to be careful this doesn't lead to market concentration with these companies, simply because the merchant doesn't think she can trust an open-source agent.

Liability is the 'unsettled' issue at the heart of this problem. Anthropic says in the prospectus that existing legal doctrines governing agency, product liability and professional negligence were 'developed in the context of human decision-making, and have not been authoritatively applied to autonomous AI systems'. How either Claude or those deploying it may be held responsible for autonomous actions is a challenge to legal doctrines. And laws are likely to vary across the world.

Insurance may be a solution. But it's limited. Anthropic warns that 'no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure'.

As India begins work towards an AI law, it must realise that AI and agents challenge the very nature of regulation: from deterministic to probabilistic, and from human-driven to autonomous outcomes that often humans can observe only after the fact. At the same time, countries that resolve some of these issues and become agent-ready can give their industry access to intelligence previously available only to the developed world.

The writer is founder, MediaNama
(Disclaimer: The opinions expressed in this column are that of the writer. The facts and opinions expressed here do not reflect the views of www.economictimes.com.)
Download
The Economic Times Business News App
for the Latest News in Business, Sensex, Stock Market Updates & More.
READ MORE
ADVERTISEMENT

READ MORE:

LOGIN & CLAIM

50 TIMESPOINTS

More from our Partners

Loading next story
Business News › Opinion › ET Commentary › Anthropicology Lesson: The future goes public
Text Size:AAA
Success
This article has been saved

*

+