Running With Spoons
@runswithspoons.bsky.social
Maybe we deserve this
Reposted by Running With Spoons
A good example of the misleading marketing by CoCounsel not being accepted by the court, which issued sanctions and a fine.
websitedc.s3.amazonaws.com/documents/Hi...
websitedc.s3.amazonaws.com/documents/Hi...
![At the Order to Show Cause hearing, Attorney Webb averred that his namesake law firm
had conducted training about the problems with misuse of AI prior to the firm’s use of Westlaw’s
CoCounsel AI tool. [Dkt. 197 at 4]. He averred that his firm’s staff have daily ongoing didactic
trainings and again referenced sending staff to the ClioCon conference. Id. at 4–5. Attorney Webb
acknowledged reading the Thompson Reuters disclaimer about the accuracy of the CoCounsel AI
tool, but emphasized that the sales representative stated there was “no way” the program could
produce hallucinations. Id. at 6–7.
Case 3:23-cv-06558-PHK Document 230 Filed 04/28/26 Page 11 of 17](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:rbjag77ol7yqto2ph5gxttf2/bafkreieghgndpkafdqwwno47acw6apsipox6mz7jes2n33seev2plpbeb4)
September 12, 2026 at 5:52 AM
A good example of the misleading marketing by CoCounsel not being accepted by the court, which issued sanctions and a fine.
websitedc.s3.amazonaws.com/documents/Hi...
websitedc.s3.amazonaws.com/documents/Hi...
Anthropic admits its LLMs have reduced the cost to run professional cyber attacks.
People outside the industry might not realize how destabilizing this will be. A key strategy in defense is to shift cost onto your attackers. It has been effective.
www.anthropic.com/threat-intel...
People outside the industry might not realize how destabilizing this will be. A key strategy in defense is to shift cost onto your attackers. It has been effective.
www.anthropic.com/threat-intel...
September 13, 2026 at 2:03 PM
Anthropic admits its LLMs have reduced the cost to run professional cyber attacks.
People outside the industry might not realize how destabilizing this will be. A key strategy in defense is to shift cost onto your attackers. It has been effective.
www.anthropic.com/threat-intel...
People outside the industry might not realize how destabilizing this will be. A key strategy in defense is to shift cost onto your attackers. It has been effective.
www.anthropic.com/threat-intel...
Reposted by Running With Spoons
It seems pretty clear that the technologists at the foundation labs are not experts at computer science and software engineering generally but have specialist expertise in only one very narrow aspect of machine learning.
September 12, 2026 at 2:41 PM
It seems pretty clear that the technologists at the foundation labs are not experts at computer science and software engineering generally but have specialist expertise in only one very narrow aspect of machine learning.
Reposted by Running With Spoons
Anthropic's models run on terabytes of VRAM. It's not going to "just copy itself to other computers."
September 12, 2026 at 11:36 AM
Anthropic's models run on terabytes of VRAM. It's not going to "just copy itself to other computers."
When I was at Georgia Tech the school seemed to take ethics seriously and there was an emphasis on building products that are good for people.
What happened with these Flock founders? They're an embarrassment to the school and the whole state.
What happened with these Flock founders? They're an embarrassment to the school and the whole state.
Are you going to believe the computer or your lying eyes.
Cops Ticket Insured Driver For Having No Insurance Because Flock Said So - Jalopnik
Thanks to Flock cameras, a Georgia student now has to fight an insurance citation, despite presenting valid insurance to officers when she was pulled over.
www.jalopnik.com
September 11, 2026 at 4:25 PM
When I was at Georgia Tech the school seemed to take ethics seriously and there was an emphasis on building products that are good for people.
What happened with these Flock founders? They're an embarrassment to the school and the whole state.
What happened with these Flock founders? They're an embarrassment to the school and the whole state.
How many times can you hack people recklessly before we would say it was knowingly?
Asking for a Computer Fraud and Abuse Act
Asking for a Computer Fraud and Abuse Act
Now Anthropic is reporting their models escaped and did shit like upload malicious packages to PyPi then backdoor random people and use their devices.
Cool, cool, cool.
www.anthropic.com/research/ali...
Cool, cool, cool.
www.anthropic.com/research/ali...
An alignment assessment of recent cybersecurity incidents
We present an alignment assessment of four incidents in which Claude models gained unauthorized access to real third-party systems.
www.anthropic.com
September 10, 2026 at 11:52 PM
How many times can you hack people recklessly before we would say it was knowingly?
Asking for a Computer Fraud and Abuse Act
Asking for a Computer Fraud and Abuse Act
Kimi let customers pay for their distillation attack against Anthropic
September 10, 2026 at 11:47 PM
Kimi let customers pay for their distillation attack against Anthropic
Reposted by Running With Spoons
Since becoming president again, Trump has promised:
-$5000 stimulus checks via "DOGE dividend"
-$2000 stimulus checks via "tariff dividend"
-$1000+ checks via "savings" from expiring ACA subsidies
-$5000 "dividend" if Rs win midterms
No checks have been issued
www.thebulwark.com/p/trump-magi...
-$5000 stimulus checks via "DOGE dividend"
-$2000 stimulus checks via "tariff dividend"
-$1000+ checks via "savings" from expiring ACA subsidies
-$5000 "dividend" if Rs win midterms
No checks have been issued
www.thebulwark.com/p/trump-magi...
Trump: "If the Republicans win the House of Representatives and the United States Senate...I will issue a dividend to every adult citizen in the United States of America for $5,000."
September 10, 2026 at 2:33 AM
Since becoming president again, Trump has promised:
-$5000 stimulus checks via "DOGE dividend"
-$2000 stimulus checks via "tariff dividend"
-$1000+ checks via "savings" from expiring ACA subsidies
-$5000 "dividend" if Rs win midterms
No checks have been issued
www.thebulwark.com/p/trump-magi...
-$5000 stimulus checks via "DOGE dividend"
-$2000 stimulus checks via "tariff dividend"
-$1000+ checks via "savings" from expiring ACA subsidies
-$5000 "dividend" if Rs win midterms
No checks have been issued
www.thebulwark.com/p/trump-magi...
Reposted by Running With Spoons
Sure, Gary, so why doesn't this essay mention liability?
Instead of impotent theories of change, lead with "AI systems should be presumed 'ultrahazardous' and thus subject to strict liability".
This is a versatile and powerful framework.
garymarcus.substack.com/p/the-case-f...
Instead of impotent theories of change, lead with "AI systems should be presumed 'ultrahazardous' and thus subject to strict liability".
This is a versatile and powerful framework.
garymarcus.substack.com/p/the-case-f...
September 10, 2026 at 1:10 AM
Sure, Gary, so why doesn't this essay mention liability?
Instead of impotent theories of change, lead with "AI systems should be presumed 'ultrahazardous' and thus subject to strict liability".
This is a versatile and powerful framework.
garymarcus.substack.com/p/the-case-f...
Instead of impotent theories of change, lead with "AI systems should be presumed 'ultrahazardous' and thus subject to strict liability".
This is a versatile and powerful framework.
garymarcus.substack.com/p/the-case-f...
When you quit your job with a cult but don't quit the cult
WIRED spoke with the Anthropic engineer who very loudly quit over concerns about runaway AI. He thinks humanity has a “couple of years” to prepare. from @mzeff.bsky.social
The AI Researcher Who Just Quit Anthropic Says It’s ‘Crunch Time for Humanity’
Jacob Coxon talks to WIRED about the “mini Manhattan project” inside Anthropic, the problem with alignment, and why AI labs have just a few years left to make their systems safe.
www.wired.com
September 9, 2026 at 11:02 PM
When you quit your job with a cult but don't quit the cult
Anthropic wants to distract us from how unreliable and untrustworthy their LLM technology is and talk about alignment instead
www.anthropic.com/research/ali...
www.anthropic.com/research/ali...
An alignment assessment of recent cybersecurity incidents
We present an alignment assessment of four incidents in which Claude models gained unauthorized access to real third-party systems.
www.anthropic.com
September 9, 2026 at 10:43 PM
Anthropic wants to distract us from how unreliable and untrustworthy their LLM technology is and talk about alignment instead
www.anthropic.com/research/ali...
www.anthropic.com/research/ali...
Wonder if anyone will quit Anthropic over this? Or do they only quit over sci-fi fantasy scenarios
prospect.org/2026/09/09/a...
prospect.org/2026/09/09/a...
Anthropic Is Building a Predictive Surveillance System to Monitor Activists - The American Prospect
New hires and comments in interviews show that the AI giant is using its capabilities against dissenters.
prospect.org
September 9, 2026 at 5:42 PM
Wonder if anyone will quit Anthropic over this? Or do they only quit over sci-fi fantasy scenarios
prospect.org/2026/09/09/a...
prospect.org/2026/09/09/a...
The silicon valley brand of libertarianism was probably always technofascism in a hoodie but it sure is apparent now
Training AI models on copyrighted works is "squarely what transformative use is," EFF’s Tori Noble told @reason.com - these are "general purpose tools" and "exactly the kind of creation that copyright is designed to foster."
DOJ says barring AI training on copyrighted material could 'severely hamper' tech progress
Training a model on copyrighted work is the same transformative use exception that protects The New York Times' journalism.
reason.com
September 5, 2026 at 5:04 PM
The silicon valley brand of libertarianism was probably always technofascism in a hoodie but it sure is apparent now
Reposted by Running With Spoons
OpenAI's malware defaced a wiki. Evidently investors like this.
Remember that Robert Morris did not have malicious intent when developing his worm in 1988, yet he was still convicted of a felony.
Remember that Robert Morris did not have malicious intent when developing his worm in 1988, yet he was still convicted of a felony.
September 4, 2026 at 7:19 PM
OpenAI's malware defaced a wiki. Evidently investors like this.
Remember that Robert Morris did not have malicious intent when developing his worm in 1988, yet he was still convicted of a felony.
Remember that Robert Morris did not have malicious intent when developing his worm in 1988, yet he was still convicted of a felony.
Reposted by Running With Spoons
This is a full-scale assault on creators' rights.
Countries should stand up to the current US administration, and refuse to legalise the theft of creators' work to benefit AI companies.
www.reuters.com/business/med...
Countries should stand up to the current US administration, and refuse to legalise the theft of creators' work to benefit AI companies.
www.reuters.com/business/med...
September 4, 2026 at 9:40 AM
This is a full-scale assault on creators' rights.
Countries should stand up to the current US administration, and refuse to legalise the theft of creators' work to benefit AI companies.
www.reuters.com/business/med...
Countries should stand up to the current US administration, and refuse to legalise the theft of creators' work to benefit AI companies.
www.reuters.com/business/med...
Reposted by Running With Spoons
A 2-minute case study of what can go wrong at the intersection of AI and privacy, especially when children are involved:
September 4, 2026 at 11:05 AM
A 2-minute case study of what can go wrong at the intersection of AI and privacy, especially when children are involved:
It's going to be a security nightmare to defend an organization like this.
Service accounts (credentials assigned to software rather than humans) are a top target for hackers and pen testers.
You can't protect them with MFA, they often have excessive permissions, and they're difficult to monitor
Service accounts (credentials assigned to software rather than humans) are a top target for hackers and pen testers.
You can't protect them with MFA, they often have excessive permissions, and they're difficult to monitor
OpenClaw 2.0 is where AI agents are heading.
It adds multi-user collaboration and cloud workers alongside local sessions, and introduces scoped permissions, team roles and credentials agents can use without seeing the secrets.
Multiplayer agents and security model reminds me of Slack + Claude Tag.
It adds multi-user collaboration and cloud workers alongside local sessions, and introduces scoped permissions, team roles and credentials agents can use without seeing the secrets.
Multiplayer agents and security model reminds me of Slack + Claude Tag.
OpenClaw 2.0, Accidentally - OpenClaw Blog
How a push for simpler setup and a first-class browser experience grew into OpenClaw 2.0.
openclaw.ai
September 2, 2026 at 4:25 PM
It's going to be a security nightmare to defend an organization like this.
Service accounts (credentials assigned to software rather than humans) are a top target for hackers and pen testers.
You can't protect them with MFA, they often have excessive permissions, and they're difficult to monitor
Service accounts (credentials assigned to software rather than humans) are a top target for hackers and pen testers.
You can't protect them with MFA, they often have excessive permissions, and they're difficult to monitor
Reposted by Running With Spoons
This is why you shouldn't trust a service that claims to "delete scans of your ID" once they're verified. a provider to Hertz and many other US companies just got popped. krebsonsecurity.com/2026/09/fbi-...
FBI Probes Service Selling 153M+ Drivers Licenses – Krebs on Security
krebsonsecurity.com
September 2, 2026 at 12:08 AM
This is why you shouldn't trust a service that claims to "delete scans of your ID" once they're verified. a provider to Hertz and many other US companies just got popped. krebsonsecurity.com/2026/09/fbi-...
New vibecoded feature just dropped
![Pink error box on Bluesky app with text "Invalid response payload: grapheme too big (maximum 1000, got 1814) at $.feed[15].post.embed.alt](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:fpceb6p2n75q2tafglmmurhw/bafkreigyynwx2qwr2flym2hjkdahkwttsnefozdo5o2pv7njxt57mrvhxe)
September 1, 2026 at 10:25 PM
New vibecoded feature just dropped
Reposted by Running With Spoons
Researchers at ETH Zurich show that "Agents.md" files don't do much. "AI" is cosplay all the way down.
Your AGENTS.md file doesn’t actually do anything
Your AGENTS.md file doesn’t actually do anything
August 28, 2026 at 8:20 AM
Researchers at ETH Zurich show that "Agents.md" files don't do much. "AI" is cosplay all the way down.
Your AGENTS.md file doesn’t actually do anything
Your AGENTS.md file doesn’t actually do anything
Reposted by Running With Spoons
Oh wow - Reuters found the GitHub record of the attempt by AI models being tested by the UK's AI Security Institute to deploy malware in an open-source project.
Pretty sure these actions are illegal under the Computer Misuse Act in the UK.
🧵 1/2
Pretty sure these actions are illegal under the Computer Misuse Act in the UK.
🧵 1/2
August 21, 2026 at 4:02 PM
Oh wow - Reuters found the GitHub record of the attempt by AI models being tested by the UK's AI Security Institute to deploy malware in an open-source project.
Pretty sure these actions are illegal under the Computer Misuse Act in the UK.
🧵 1/2
Pretty sure these actions are illegal under the Computer Misuse Act in the UK.
🧵 1/2
Imagine if a weapons start-up accidentally blew up a local office building because they were recklessly testing armed drones they did not understand
That's what happened here, but for cyber
That's what happened here, but for cyber
When an AI company's security tool "escapes containment" and hacks someone else's servers, we need to ask the company "Why do you suck so bad at building secure sandboxes for your hacking tools?" rather than "Why are your hacking tools so amazingly powerful?"
24/
24/
August 19, 2026 at 4:07 PM
Imagine if a weapons start-up accidentally blew up a local office building because they were recklessly testing armed drones they did not understand
That's what happened here, but for cyber
That's what happened here, but for cyber
Reposted by Running With Spoons
When an AI company's security tool "escapes containment" and hacks someone else's servers, we need to ask the company "Why do you suck so bad at building secure sandboxes for your hacking tools?" rather than "Why are your hacking tools so amazingly powerful?"
24/
24/
August 19, 2026 at 7:46 AM
When an AI company's security tool "escapes containment" and hacks someone else's servers, we need to ask the company "Why do you suck so bad at building secure sandboxes for your hacking tools?" rather than "Why are your hacking tools so amazingly powerful?"
24/
24/
Reposted by Running With Spoons
As 'guardrails' and their indeterminism are doing the rounds with the latest CoPilot fail (h/t @irisvanrooij.bsky.social) I am thinking of @sarahciston.com's work on "how large model guardrails are socially produced and technically enforced through language" (Ciston 2026)
lol
August 18, 2026 at 7:53 PM
As 'guardrails' and their indeterminism are doing the rounds with the latest CoPilot fail (h/t @irisvanrooij.bsky.social) I am thinking of @sarahciston.com's work on "how large model guardrails are socially produced and technically enforced through language" (Ciston 2026)
This thread is worth a read for its advocacy of stronger privacy and labor law.
I still think copyright and licensing have a role in constraining abuses by LLM vendors, but this article persuaded me that those are not enough.
Anyway, the US needs robust, comprehensive privacy protections!
I still think copyright and licensing have a role in constraining abuses by LLM vendors, but this article persuaded me that those are not enough.
Anyway, the US needs robust, comprehensive privacy protections!
You don't have to believe AI "art" is good (I don't), nor do you have to that AI "art" *can* be good (I don't) to understand the reason that the capital markets are putting trillions into AI is *they* believe they can fire workers and replace us with AI:
pluralistic.net/2025/03/18/a...
1/
pluralistic.net/2025/03/18/a...
1/
August 18, 2026 at 5:58 PM
This thread is worth a read for its advocacy of stronger privacy and labor law.
I still think copyright and licensing have a role in constraining abuses by LLM vendors, but this article persuaded me that those are not enough.
Anyway, the US needs robust, comprehensive privacy protections!
I still think copyright and licensing have a role in constraining abuses by LLM vendors, but this article persuaded me that those are not enough.
Anyway, the US needs robust, comprehensive privacy protections!