Claude’s confessional: Anthropic courts the pope while dropping Opus 4.8
Anthropic is currently pulling off the tech industry’s most ambitious identity crisis. One day, co-founder Chris Olah is standing in the Vatican, admitting that AI labs are basically powered by pride and "geopolitical pressure." The next, the company is shipping Claude Opus 4.8, promising that this version is finally—mostly—done with the lying.
It’s a fascinating, if slightly surreal, bit of brand positioning. By leaning into the "mysterious" and "unsettling" nature of their own tech, Anthropic is trying to be the adult in the room. In the spirit of clarity, it’s a masterclass in brand safety: if you can’t guarantee the robot won’t hallucinate, at least get the Pope to bless the "discernment" process.
Olah’s remarks at the Vatican presentation of the encyclical Magnifica humanitas were surprisingly raw for a C-suite executive. He didn't just talk about math; he talked about "fictional characters brought to life" and internal states that mirror human "joy, satisfaction, fear, and grief." It sounds like the plot of a B-tier sci-fi movie, but for Olah, it’s a reason to beg for outside critics who aren't blinded by the industry's profit-first incentives.
While the theology was happening in Rome, the engineering team in San Francisco was busy shipping the actual hardware. Opus 4.8 is here, and the headline feature is "honesty." Apparently, the previous models were a bit too eager to jump to conclusions—tech speak for "making things up to look smart." Anthropic claims this update is four times less likely to let bugs pass unremarked, which is a relief for anyone tired of gaslighting by their IDE.
We also got a new "effort control" toggle, which is essentially a slider for how much you want Claude to actually think. On "Max" or "Extra" effort, the model will burn more tokens and spend more time ruminating on a task. It’s a pragmatic move, but let’s be real: it’s also an admission that the default settings were sometimes just phoning it in.
For the power users, "dynamic workflows" are the real star. This allows Claude to spin up hundreds of parallel subagents to tackle massive codebase migrations. Looking at this practically, it’s the beginning of the "autonomous coworker" era they’ve been teasing. Whether you actually want hundreds of subagents touching your legacy code is a question of faith—one the Pope might not be able to help with.
Between the Vatican visit and the Glasswing cybersecurity safeguards, Anthropic is desperate to be seen as the "safe" lab. They’re even hinting at a "Mythos-class" model that is so powerful it needs extra protection before it hits the public. It’s a lot of hype wrapped in a shroud of humility. For now, we have a smarter, slightly more honest Opus 4.8 that can "think harder" if you pay for it.
Sources: Chris Olah's Remarks at the Vatican, Introducing Claude Opus 4.8.
