Anthropic is different because they're committed to safety.

[fritzadalis@infosec.exchange]

@mttaggart
Fixed:
Remember kids, any promise from corpos is a bald-faced lie.

[theorangetheme@en.osm.town]

@mttaggart Just in time for them to make a sale to the Pentagon, which wants fewer safeguards and also threatened Anthropic if they didn't drop the existing safeguards they do have (lol): https://www.theguardian.com/us-news/2026/feb/24/anthropic-claude-military-ai

[rachel@transitory.social]

@mttaggart@infosec.exchange

Thinking about how Google quietly removed any mention of "Don't be evil"

[mttaggart@infosec.exchange]

@theorangetheme If they fold on that one, they deserve all the excoriation they'll get. You can't go "We won't let you build a murderbot with our AI" and then later go "Actually murderbots are fine."

[theorangetheme@en.osm.town]

@mttaggart I'm not even sure what the military would do with Claude anyway. They have the same number of guns as before, but I guess now they can write slop reports faster?

[theorangetheme@en.osm.town]

@mttaggart Cynical: a human can still show up and violate my rights more effectively than AI.

[jrdepriest@infosec.exchange]

@theorangetheme @mttaggart

Autonomous death robots for plausible deniability during any future Nuremberg-like trials.

[bruce@darkmoon.social]

@mttaggart

Huh. They went from zero to evil in 0.05 googles.

[flashmobofone@mastodon.art]

@mttaggart Yeah, gathered that when they capitulated to incompetent fascist psychopaths.

[hugo@social.treehouse.systems]

@mttaggart

Anthropic’s chief science officer Jared Kaplan told TIME in an exclusive interview. “We didn't really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.”

Kaplan, the Anthropic executive and co-founder, denied the company’s decision to change course was a capitulation to market incentives as the race for superintelligence accelerates.

Amazing. You can literally state this as
"While we're doing this because other people in the market can move faster if they don't also have this restriction, meaning the market is providing incentives to remove this restriction, we categorically deny that removing this restriction has anything to do with market incentives."

[wyatt_h_knott@vermont.masto.host]

@mttaggart do these people not understand what a pledge is?

[mttaggart@infosec.exchange]

@Wyatt_H_Knott They understand that it's great marketing, until it gets in the way of business

[wyatt_h_knott@vermont.masto.host]

@mttaggart Someone should tell them that we still haven't forgiven google for giving up on their pledge.

[riley@toot.cat]

@mttaggart Because the first law of aibotics is An AIbot may not injure its grifters' quarterly cash flow or, by inaction, allow said cash flow to come to harm before the quarterly report is out.

[negativek@infosec.exchange]

@mttaggart I'm an optimist. Sometimes they believe what they say and then are fired.

Or they just become a different person. A person that wants money more.

[bltpizza@mastodon.social]

@mttaggart AI WILL IMPRISON US.

[gimulnautti@mastodon.green]

@mttaggart Didn’t the US DOD threaten them with contract loss if they don’t agree to their models being used to kill people?

I would guess they would have to go pretty deep into the core of their reasoning training material to make the models not refuse hurting humans. 🤔

[gimulnautti@mastodon.green]

@Wyatt_H_Knott @mttaggart Conclusion:
Capitalism literally kills people for profit.

[dryak@mstdn.science]

@theorangetheme @mttaggart Wait until some brilliant mind decides to wires those triggers to some agentic AI bot...

[cppguy@infosec.space]

@theorangetheme @mttaggart

"OK, Claude. Whenever you see an enemy plane, I want you to shoot it down without asking me first."

"OK, Claude. Whenever you see a Palestinian, I want you to highlight that person in my HUD. This applies especially if the person is trying to hide from me."