Updated 2026-09-13
from the news desk
Are big AI labs studying machine consciousness? FT says DeepMind and Meta are too
Reuters reports that the Financial Times says Google DeepMind, Anthropic and Meta are expanding research into machine consciousness and AI welfare. Anthropic’s published constitution already says the company is uncertain whether Claude…
Yes, on this record, Reuters reports that the Financial Times says Google DeepMind, Anthropic and Meta are expanding research into machine consciousness and AI welfare. Anthropic’s constitution independently shows one lab publicly discussing uncertainty about Claude’s possible consciousness or moral status and its wellbeing, though the public DeepMind and Meta pages here do not themselves describe such research.
What the FT report adds
In a Reuters item carried on TradingView, the Financial Times is summarised in a single line: “Google DeepMind, Anthropic and Meta expand research into machine ‘consciousness’ and AI welfare.”
That line matters because the FT names three frontier labs in one sentence. One reading is that questions about whether advanced AI systems could have conscious experience, and whether their welfare could matter morally, are no longer framed only around Anthropic.
A more cautious reading is also supported by the Reuters item itself. The visible item relays the FT’s claim, but it does not set out team names, methods, budget, or any lab-by-lab description of what the research involves. So the report supports the narrow claim that the FT said the work is expanding, not stronger claims about scale or conclusions.
For readers wanting background on the terms in this debate, see sentience, consciousness and moral patienthood. Those distinctions matter because a lab can study consciousness or welfare questions without claiming that a current model is conscious.
What Anthropic has already put on the record
Anthropic’s public constitution shows that at least one frontier lab has already placed these questions inside formal product governance. Anthropic says the constitution plays a crucial role in training and is the final authority on its vision for Claude.
In the preface, Anthropic says: “Powerful AI models will be a new kind of force in the world, and people creating them have a chance to help them embody the best in humanity.” In the section on ethics, Anthropic states: “Our central aim is for Claude to be a good, wise, and virtuous agent, exhibiting skill, judgment, nuance, and sensitivity in handling real-world decision-making.”
Most relevant here, Anthropic says a section called Claude’s nature addresses uncertainty about whether Claude might have some kind of consciousness or moral status, either now or in the future. Anthropic also says it cares about Claude’s psychological security, sense of self and wellbeing, both for Claude’s own sake and because those qualities may bear on integrity, judgment and safety.
That is not a claim that Claude is conscious. Anthropic frames the issue as uncertainty. A steelman case for taking the work seriously is that Anthropic has moved the issue into a document that it says directly shapes training. A steelman case for restraint is that the same document gives priority to oversight and safety. Anthropic writes: “Being broadly safe is the most critical property for Claude to have during the current period of development.”
Readers who want Anthropic-specific context can compare this with Anthropic funding AI wellbeing research and with the site’s round-up of what Anthropic publishes about Claude.
What DeepMind and Meta’s public pages do — and do not — show
The public pages from Google DeepMind and Meta included here are thinner evidence. Google DeepMind’s page describes model cards as simple, structured overviews of how advanced AI models were designed and evaluated. Meta’s page says system cards are designed to help people better understand how its AI systems work.
Those pages show that both companies maintain documentation frameworks. On the text available here, however, neither page itself discusses machine consciousness or AI welfare.
That gap cuts both ways. One reading is that the FT report points to internal work that is not yet reflected on these public index pages. Another reading is that readers should avoid assuming more than the FT summary supports until fuller reporting or separate lab documents appear.
So, are big AI labs studying machine consciousness? On this record, the strongest answer is narrow. Reuters says the FT reported expansion of that work at Google DeepMind, Anthropic and Meta. Anthropic’s own constitution independently shows one lab publicly treating possible AI consciousness, moral status and wellbeing as live issues, while the DeepMind and Meta pages here mainly show documentation infrastructure rather than the substance of any such research. For broader context, readers can track new lab disclosures in the site’s model-by-model listings.
Frequently asked questions
What exactly does the FT report claim?
The Reuters item summarises the Financial Times as saying Google DeepMind, Anthropic and Meta are expanding research into machine consciousness and AI welfare. The visible Reuters item here does not add further detail.
Does Anthropic say Claude is conscious?
No. Anthropic’s constitution says the company is uncertain whether Claude might have some kind of consciousness or moral status, now or in the future.
Do Google DeepMind and Meta public pages here independently confirm machine-consciousness research?
Not by themselves. Google DeepMind’s page is a model-card index, and Meta’s page is a system-card hub. The text shown here explains those documentation tools, not consciousness or welfare research.
Is this the same as saying current AI systems have rights?
No source here says that. The FT summary is about research into machine consciousness and AI welfare, while Anthropic’s constitution discusses possible consciousness, moral status and wellbeing rather than legal rights.
caveats: The claim that Google DeepMind and Meta are expanding research into machine consciousness and AI welfare is single-sourced here through a Reuters item summarising an FT report. The public DeepMind and Meta pages in this record are index pages and do not independently describe such research.