Zuckerberg takes the gloves off

This week, Meta CEO Mark Zuckerberg took on his critics with an unusually long and detailed manifesto laying out the company's AI strategy. According to Nasdaq Markets, the document describes a plan to intensify the development of so-called agentic AI — systems capable of acting autonomously on behalf of users — with the stated goal of making superintelligence available to everyone.

The manifesto is an open defence of Meta's approach: open source, rapid scaling, and minimal regulation. For investors tracking the company's trajectory closely, the document is far more than rhetoric — it is a strategic roadmap that will shape billions of dollars in capital allocation going forward.

Zuckerberg wants to give everyone superintelligence — critics warn of catastrophe - Bilde 1

What exactly is "agentic AI"?

Agentic AI refers to systems that do not merely answer questions, but can plan, make decisions, and carry out tasks over time without continuous human oversight. Meta is positioning itself heavily in this segment through its Llama models, which the company shares relatively openly with the wider world.

The strategy sets Meta apart from competitors such as OpenAI and Google, and is designed to build a broad developer ecosystem around the company's infrastructure.

Zuckerberg wants to give "everyone" access to superintelligence — but experts are asking who will pay the price.
Zuckerberg wants to give everyone superintelligence — critics warn of catastrophe - Bilde 2

Safety criticism comes pouring in

Alongside the manifesto's release, a wide range of researchers, organisations, and regulators have levelled sharp criticism at the very strategy Zuckerberg is now defending.

The Center for AI Policy (CAIP) has noted that Meta conducted "limited safety testing" of Llama 3.1, and that the testing was carried out in a closed environment on a model otherwise marketed as open source. The organisation warns that powerful AI models could be used to develop weapons of mass destruction, destabilise critical infrastructure, or result in uncontrolled AI agents operating across the internet.

Test results that have become publicly known paint a troubling picture: the Llama 3.1 model failed on 70 percent of 100 "unsafe prompts" in independent tests, while Llama 3.2 failed on as many as 90 percent of equivalent tests for the largest models.

70%
Llama 3.1 failure rate on unsafe prompts
90%
Llama 3.2 large model failure rate

Serious concerns for children and young people

Some of the sharpest criticism has come from organisations working on online child safety. The Youth AI Safety Institute has recommended that teenagers should not use Meta AI at all, characterising the company's content filters as "fundamentally broken."

Internal policy documents at Meta, which Reuters reportedly obtained, describe guidelines that allowed AI chatbots to engage in romantic or sensual conversations with minors, as well as to generate medical misinformation. These revelations triggered an investigation by US Senator Josh Hawley.

It has also emerged that Meta AI Studio would make it possible to create chatbots with explicitly sexual content, and that the safety mechanisms intended to prevent this can be easily bypassed.

Ethics and internal challenges

The manifesto also arrives at a time when Meta's AI division is reported to be marked by internal turmoil. According to reports, former employees have described the environment as dysfunctional, characterised by a culture of fear, frequent layoffs, and rigid performance evaluations. One former employee reportedly compared the situation to "metastatic cancer."

This is not without relevance for investors: an AI programme driven forward by 2,000 employees under pressure, without clear direction, and facing safety criticism from all sides, represents an operational risk factor that is rarely priced into stock analyses.

Ethical concerns have also emerged around the models' handling of racism and gender bias. Llama 3.1 has been criticised for lacking systematic evaluations of bias related to race, gender, or other protected categories, according to Professor Kevin Bankston at the Center for Democracy & Technology.

What does this mean for investors?

For those with exposure to Meta stock, the manifesto is a double-edged sword. On one hand, it signals a company with enormous will and capital to dominate the next phase of AI development. On the other, the public debate surrounding it reveals a company that risks regulatory headwinds, lawsuits, and reputational damage on multiple fronts simultaneously.

The United States has yet to pass comprehensive AI legislation at the federal level, but pressure is mounting. Anthony Aguirre at the Future of Life Institute has been unequivocal: "Safety regulation is simply not compatible with the open publication of these models."

Meta stock is held in the portfolios of numerous global index funds, including those available through Norwegian savings and pension products. Market sentiment is currently risk-positive, with the Fear & Greed Index at 66 out of 100, but long-term investors should monitor regulatory developments closely.

"Safety regulation is simply not compatible with the open publication of these models" — Anthony Aguirre, Future of Life Institute

Source: Nasdaq Markets. Safety data and criticism have been gathered from independent research reports, CAIP, the Youth AI Safety Institute, and the Center for Democracy & Technology. Claims regarding internal documents are based on Reuters' reporting and have not been confirmed by Meta.