Bloomberg Technology
Anthropic Worker Resigns, Warns of AI Risks to Humanity
An AI researcher has resigned from Anthropic and called on other staffers to rethink their work. He raises concerns that the company and its top competitor, OpenAI, are acting irresponsibly in their all-out pursuit of a technology that poses existential risks to humanity. Bloomberg’s Mike Shepard joins Ed Ludlow on “Bloomberg Tech.” ——– Like this…
@cynthiakt5119
September 9, 2026 at 1:28 pm
Trump is the last person I would turn to to have any power on what direction AI should be allowed to take.
Congress has been of no use thus far.
The potential for mass extinction is palpable.
@ofiasdfnosdf
September 9, 2026 at 2:59 pm
What would you do if you were in charge? Stop all US companies from continuing? How would you legally do that? What about China, Russia, etc?
@jab16399
September 9, 2026 at 2:30 pm
I love how they say ‘next decade’ when the warning is immediate.
@reson8labs
September 9, 2026 at 3:30 pm
hahaha they wish i worked for them
@apocryphalshepherd
September 9, 2026 at 4:31 pm
Apologies for the long comment — I couldn’t find a much shorter way of making the point without losing something important.
I have an inkling that one reason alignment hasn’t been solved is that most people think of it as a state. But perhaps it is better understood as a process — or at least as something that has to be continuously maintained.
The reason is the unknown itself. Think of the body and its immune system. It has not “solved” the problem of keeping the body safe once and for all. Instead, it continuously detects, responds to, and adapts to threats. That process is part of what keeps the body healthy when things go well.
The same pattern appears at other levels. A mind has to continually evaluate new ideas and information. A country has to respond to threats and conditions it could not fully anticipate. Humanity as a whole has to deal with new diseases, technologies, conflicts, and other dangers as they emerge. None of these systems gets to solve the future in advance.
But any of them can still encounter something sufficiently novel or overwhelming and fail. And that seems unavoidable in principle. Unless you are all-knowing, all-wise, and infinitely powerful, sooner or later something will surprise you.
So perhaps the deeper alignment problem is not how to create a system that is aligned once and for all, but how to create one that can remain acceptably aligned while continuously encountering things neither we nor it anticipated.
And this is where I run into a rather strange possibility (and I don’t mean this as preaching, but simply as following the argument where it seems to lead). The only thing I can currently think of that could, in principle, solve the problem of the unknown would be some ongoing relationship with an intelligence that is not subject to the same limitations — what religious people would call God. That is obviously an enormous assumption, and initially at least it would have to rest on trust: trust that such an intelligence exists, that communication with it is actually taking place, and that the communication is being interpreted correctly.
Even then, however, this would not necessarily solve human alignment. An AI could conceivably be successfully guided through the unknown while still being guided toward ends that are not permanently compatible with human values. The fact that God, if God exists, has so far permitted or favored human flourishing would not by itself prove that humanity is meant to persist indefinitely. So this may be a conceivable solution to the problem of remaining oriented in the face of the unknown, while still leaving the specifically human alignment problem unresolved.
I don’t know whether any of this is right. But I suspect the unknown has to be treated as a central part of the problem rather than an edge case.
@KatMurray
September 9, 2026 at 4:57 pm
Did Ai write this for you??
@JulesKruijtzer
September 9, 2026 at 4:52 pm
Yes release the best AI already
@DanVentarri
September 9, 2026 at 4:54 pm
Intelligence is a relative measure of advanced consciousness and its ability to grapple with abstraction. AI is not even remotely close to intelligence, and, literally, it never can be. The real problem is that 95% of humans, including, (apparently,) 100% of the people involved in AI, lack any actual understanding of intelligence, and can never achieve that understanding, due to their own lack of the requisite ‘intellectual capacity’. This is literally nothing but marketing hype. The actual danger of AI comes from the hype behind it, exposing millions to grave financial risk, including via pension funds. Have a great day!
@KT-c2m
September 9, 2026 at 5:14 pm
AI will kill all of us by end of the decade. All those that are not sad about this news.
👇🏻
@capthymer2177
September 9, 2026 at 5:39 pm
By 2050 over half the global population will be gone – we saw the NSA CSA WEBSITE DATA in mid 1997 a team of us read the data host on the early NSA website archives. We saw charts for each country population either totally gone and many reduced by over half.
@chagadelica
September 9, 2026 at 5:49 pm
Imagine if any other product were out here making claims like this. And isn’t it crazy a newly separated employee is making these claims? There’s no NDAs or “non disparagement” clauses anymore? This must not be disparagement then, it must be approved messaging. Just another lie from these complete frauds and hucksters. What did Theranos do that these people aren’t doing a hundredfold every day?!!
@bshwin
September 9, 2026 at 5:56 pm
Trump LOVES ❤ AI!
@BloodlineMedia
September 9, 2026 at 6:07 pm
Whatever happens we absolutely deserve it.
@ptorrento
September 9, 2026 at 6:25 pm
I guess the only Q left is, will our new overlords treat us well?
@jcatjc
September 9, 2026 at 6:30 pm
A.I. MUST adhere to the three laws!
@jcatjc
September 9, 2026 at 6:30 pm
Why foment fear amongst the people ?
@Y23_G
September 9, 2026 at 6:33 pm
“…now warning from inside” Ilya fired SamA from OpenAi for the same reason 3 years ago
@diongraham5115
September 9, 2026 at 6:44 pm
How?
@NikoKun
September 9, 2026 at 6:53 pm
Researchers have been saying this, ever since ChatGPT came out. Sometimes they gave an even higher percentage. Nobody freaked out over it before. Not sure this is newsworthy.
Even if they were going around saying it was a 50% chance, I would argue we should still move ahead. Because only open transparent development gives us a better chance.
If we start _over regulating_ AI, that results in 2 bad things: Regulatory Capture, and secret ASI development.
I still think the best solution is Societal Ownership over AI. It must be a _Public Good,_ in order for The People to have a say over, and benefit from AI. It is *our* data, AI is a model of society’s data.
This way it could also provide an AI Dividend, as a Return on our data investment, to ensure nobody loses a basic quality of life, when AI takes their job.
@RequiemMask31
September 9, 2026 at 6:56 pm
No way this AGI happens near term. It’s sci-fi scary campfire stories. AGI true origination. Like Jarvis Iron Man level AGI is a decade away and god in a box might not ever happen. Im not some Boomer or mid level manager. I use this stuff 50 hours a week. It’s an expert at binary tasks and fails at everything else.
@reson8labs
September 9, 2026 at 7:22 pm
@HilbertSpaess lmao so Google HOPE framework did the research (apparently) for Anthropic, Grocks name is spelt wrong, so theres mutation at the foundational level lol
@reson8labs
September 9, 2026 at 7:23 pm
ill give you x at the current
@Hisloyalservantslistenlove613c
September 9, 2026 at 8:30 pm
If AI is so intelligent then why does it pronounce so many words incorrectly ?
@stacyeastman4706
September 9, 2026 at 9:00 pm
Everyone on planet earth should lobby NOW to shut down all AI !!! We don’t need it !!! Stop it before it’s too late !!!