Anthropic, Open-Weight Models and the Trust Debate

When Tanya Dua and the LinkedIn News team reached out to me to share my thoughts about Anthropic refusing to sign a major AI industry letter, it brought up a huge debate that affects everyone using AI today.

What happened

Almost every big tech company (Google, Microsoft, IBM, NVIDIA, Palantir Technologies, Hugging Face and others) signed a letter telling the U.S. government not to restrict "open-weight" AI models - AI programs whose core inner workings are free for anyone to download and modify.

Anthropic's CEO, Dario Amodei, chose not to sign.

His reasoning:

• Permanent risks: once an AI model is opened up to the public, anyone can easily remove its safety filters. If bad actors get hold of it, you can't "take it back."
• Smarter rules, not bans: Amodei says he doesn't want open AI banned. Instead, he wants smarter rules - like keeping super-advanced computer chips away from dangerous groups and requiring safety checks before any major AI is released.

Why many aren't buying it

• Looks like protectionism: because companies like Meta, Microsoft, Google and OpenAI signed the letter, Anthropic stands out. Critics say safety fears are being used as an excuse to hurt free competition and push people toward paid services.
• Hurting trust: many software developers rely on free, open AI models to build customized tools without paying big tech companies monthly subscription fees. They see Anthropic's stance as an attempt to lock down the market.

My thoughts

This whole debate boils down to one simple question: does making AI free and open help the "good guys" more than the "bad guys"?

• The "coalition" side: if millions of people have access to the code, developers worldwide can quickly spot safety flaws, fix them, and build defenses together.
• The "cautious" side: a bad actor only needs to break the rules once to cause major damage, while building defenses takes a lot of time.

Until we have a clear answer, Anthropic's stance will keep dividing people - some see it as responsible caution, while others see it as unfair gatekeeping.

A question for builders: does an AI company's stance on open vs. closed models change whether you use their products, or do you just pick whatever tool is cheapest and fastest?