OpenAI Tightens Security While Anthropic Eases Model Constraints

In one line: According to The Register, two leading AI labs appear to be moving in opposite directions on security and autonomy.

Key points

• OpenAI is reported to be adding a security layer referred to as "Astra" to its systems.
• Anthropic, by contrast, is said to be loosening the constraints ("leash") on its "Fable" model.
• The contrast is notable because the two moves were reported at roughly the same time.

Why it matters

Tightening security and widening autonomy are two poles of frontier-model operation. Two flagship labs choosing opposite directions highlights how each is striking a different balance between safety and capability.

Read more

• OpenAI pledges to add Astra security as Anthropic loosens Fable's leash — The Register