TLDR Tech / InfoSec
Anthropic's Safety Superpower
Ben Thompson argues that Anthropic's safety messaging aligns a little too neatly with its commercial interests, walking through the release of Fable (a guardrailed cut of the Mythos Preview model the company had deemed too dangerous to ship) alongside its cybersecurity capabilities, the US government's export-control directive suspending access to Fable 5 and Mythos 5 for foreign nationals after a reported jailbreak, and a 30-day data-retention reversal that scrapped zero-retention guarantees even for enterprise plans. He singles out the launch-day plan to silently degrade Fable for frontier LLM development work, later walked back to a disclosed handoff to Opus 4.8, as proof that Anthropic both can and will quietly steer its models toward its own policy preferences. The throughline he draws is that a lab convinced it alone should build frontier AI, and by extension, should hold final say over AI generally, ends up framing every self-serving move as a safety necessity.
Read the full story