/Tech2h ago

Stratechery's Ben Thompson argues Anthropic's recursive self-improvement safety report is a strategic tactic to justify limiting model access

Gary Marcus endorsed the analysis, citing competitive positioning.

821K70182115.8K
Steven Sinofsky
kache
Gary Marcus#156
Original post
Gary Marcus@GaryMarcus#156inTech

Echoes what I posted about Anthropic last week.

David Sacks@DavidSacks

If you were wondering what the "pause" was all about, Ben Thompson @stratechery has an interesting theory:

"Late last week the Anthropic Institute released a new safety report warning about the danger of recursive self-improvement... I don’t think the timing is a coincidence. This is a company and leadership that has been honing safety-and-scaremongering-as-marketing-tactic ever since Amodei led the charge to close source OpenAI models because GPT-2 was too dangerous; it’s always fun to see the evolution of tactics, capabilities, and goals, and in this case publishing a widely-discussed report the week before you cite it to silently degrade your offering for potential competitors is impressive."

10:34 AM · Jun 11, 2026 · 6.2K Views
Sentiment

Many users accused Anthropic of timing its safety report as anti-competitive marketing or a Trojan horse for regulations, viewing the move as sneaky and untrustworthy.

Pos
17.6%
Neg
82.4%
18 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS3.4KLIKES20
Gary Marcus@GaryMarcus

You really should read my substack @DavidSacks (and others).

I laid out basically the same argument last week, and showed how they had been following the same strategy — which I called scare, hype, and release — since 2019.

https://open.substack.com/pub/garymarcus/p/the-revenge-of-claude-mythos?r=8tdk6&utm_campaign=post-expanded-share&utm_medium=web

lots of other stuff i published there first, like this in January 2024, extremely relevant now:

2hViews 3.4KLikes 20Bookmarks 4
BOOKMARKS8
Lost Function@thelostfunction

“Of all tyrannies, a tyranny sincerely exercised for the good of its victims may be the most oppressive. It would be better to live under robber barons than under omnipotent moral busybodies. The robber baron's cruelty may sometimes sleep, his cupidity may at some point be satiated; but those who torment us for our own good will torment us without end for they do so with the approval of their own conscience.”

—C.S. Lewis

2hViews 1.4KLikes 20Bookmarks 8
RETWEETS70
David Sacks@DavidSacks

If you were wondering what the "pause" was all about, Ben Thompson @stratechery has an interesting theory:

"Late last week the Anthropic Institute released a new safety report warning about the danger of recursive self-improvement... I don’t think the timing is a coincidence. This is a company and leadership that has been honing safety-and-scaremongering-as-marketing-tactic ever since Amodei led the charge to close source OpenAI models because GPT-2 was too dangerous; it’s always fun to see the evolution of tactics, capabilities, and goals, and in this case publishing a widely-discussed report the week before you cite it to silently degrade your offering for potential competitors is impressive."

2hViews 135.1KLikes 1.2KBookmarks 237
REPLIES2
Steven Sinofsky@stevesi

If this marketing is successful there is only one endpoint:

A very small number of models+companies (perhaps just one) become too big to fail while at the same time under the auspices of the "experts" determine what input/output models are permitted/capable of. This includes stifling alternative commercial and open source technology.

At the same time—and as a result of the above—another exceedingly small set of "trusted" parties are permitted access to additional capabilities above and beyond whatever everyone else has. This will of course happen absent any oversight or transparency.

This is the worst aspect of the deeply Orwellian evolution of the telephone, radio, and television before deregulation, cable, and the internet, or as banking and travel remain today.

The asymmetry is not a problem or bug but THE feature.

David Sacks@DavidSacks

If you were wondering what the "pause" was all about, Ben Thompson @stratechery has an interesting theory:

"Late last week the Anthropic Institute released a new safety report warning about the danger of recursive self-improvement... I don’t think the timing is a coincidence. This is a company and leadership that has been honing safety-and-scaremongering-as-marketing-tactic ever since Amodei led the charge to close source OpenAI models because GPT-2 was too dangerous; it’s always fun to see the evolution of tactics, capabilities, and goals, and in this case publishing a widely-discussed report the week before you cite it to silently degrade your offering for potential competitors is impressive."

1hViews 2KLikes 11Bookmarks 5
kache@yacineMTB

"this case publishing a widely-discussed report the week before you cite it to silently degrade your offering for potential competitors is impressive."

lmfao

David Sacks@DavidSacks

If you were wondering what the "pause" was all about, Ben Thompson @stratechery has an interesting theory:

"Late last week the Anthropic Institute released a new safety report warning about the danger of recursive self-improvement... I don’t think the timing is a coincidence. This is a company and leadership that has been honing safety-and-scaremongering-as-marketing-tactic ever since Amodei led the charge to close source OpenAI models because GPT-2 was too dangerous; it’s always fun to see the evolution of tactics, capabilities, and goals, and in this case publishing a widely-discussed report the week before you cite it to silently degrade your offering for potential competitors is impressive."

1hViews 2.4KLikes 17Bookmarks 0
Samuel Hammond 🦉@hamandcheese

@DavidSacks @stratechery Does this imply you and Ben don't think RSI is concerning per se?

Worth checking your assumptions whenever Gary Marcus agrees with you

2hViews 382Likes 8
Brandon Brooks@OfficialBBrooks

@DavidSacks @stratechery I think it’s simpler than this, they want a government deal and if they make it seem like it’s so powerful the government is forced to use it, they win.

2hViews 644Likes 4Bookmarks 1
prinz@deredleritt3r

@DavidSacks @stratechery It should be noted that OpenAI expressed its support this week for establishing an international organization that could "slow frontier development when needed".

In other words, these views are not coming only from Anthropic, although Anthropic is expressing them stronger.

2hViews 562Likes 8
0xEngineer@0xEngineers_

@DavidSacks @stratechery why do you hate anthropic so much? I don't see you having soo much hate for openai.. why?

2hViews 261
Gary Marcus@GaryMarcus

@davidmanheim @DavidSacks @zerohedge @stratechery they were terrified of GPT-2. definitely into boy who cried wolf territory.

2hViews 72Likes 1
Perry E. Metzger@perrymetzger

@hamandcheese @DavidSacks @stratechery We've had recursive self improvement for, what, at least 250 years? It's not a problem. We've always used computers and software to design new computers and software, and even now AI companies use AI to help build better AIs. So what?

2hViews 54Likes 2
FINMAN@erikfinman

@DavidSacks @stratechery This is exactly why I don’t trust most AI safety research. They’re manufacturing panic to justify more regulations that slow down innovation.

2hViews 291Likes 4
Make Money@trade2023money

@DavidSacks @stratechery @grok what this mean? especially to the Anthropic CEO's new capabilities and releases with Fable 5?

2hViews 216
David Manheim@davidmanheim

@GaryMarcus @DavidSacks @zerohedge @stratechery It's not quite "scaremongering" if they believe it. So yeah, it's possible that Anthropic is overhyping the threat from AI. It's also possible there are catastrophic risks from AU, and they either have less risk tolerance or longer time horizons for what they care about than you.

2hViews 57Likes 1
Crypto James@milkkyway01

@DavidSacks @stratechery Is Dario the true villain? We love to go after Bezos, Elon, even Sam, but the consistent drum beat of exaggerations, doomerism, anti-open source has me leaning “yes”. I’m open to changing my mind if I see new evidence to the contrary.

2hViews 454Likes 3
Gary Marcus@GaryMarcus

@davidmanheim @DavidSacks @zerohedge @stratechery which is not to say there aren’t real fears.

but what would they should have said is “don’t panic; we have this new thing; it needs attention, but we will do our best to fix it and probably can. still the broader issues very much need attention.”

2hViews 28Likes 1
David Manheim@davidmanheim

@GaryMarcus @DavidSacks @zerohedge @stratechery But you agree they were actually scared, right? Because their actions make at least as much sense, if not more, if they are actually scared, and the crying wolf is oversensitivity or much lower risk tolerance. (Rather than strategic dishonesty, as was implied.)

2hViews 24Likes 1
Edon Derguti@edon_d

@DavidSacks @stratechery One of your best reads of the industry was the Anthropic strategy since the beginning.

2hViews 541Likes 2
Harlan Stewart@HumanHarlan

@DavidSacks @stratechery Hard to say what Anthropic’s motivations are but it is true that half of all AI researchers think there are double digit odds that the technology will cause human extinction

2hViews 150Likes 3
Load more posts