Anthropic ditches its core safety promise in the middle of an AI red line fight with the Pentagon

Key Takeaway: Anthropic is replacing its once-firm safety guardrails with a more flexible, nonbinding framework—arguing that strict self-restraint can backfire in a fast-moving AI race—while simultaneously resisting Pentagon pressure to loosen red lines around weapons and surveillance.

More Insights:

• Anthropic says its old Responsible Scaling Policy could limit competitiveness and failed to spark a “race to the top” in the industry.

• A major change: it removed language about pausing training if model capabilities outpace the company’s ability to keep them safe.

• The new approach is framed as “public goals” with regular reporting, not hard commitments—intended to be iterated over time.

• The shift lands amid a Pentagon standoff: Anthropic reportedly won’t budge on AI-controlled weapons or mass domestic surveillance.

• The Defense Department allegedly threatened contract loss and “blacklist”-style penalties, intensifying the stakes.

Why it matters: Safety commitments are only as strong as the incentives behind them—once competition and national security money enter the picture, even “values-first” labs start redefining what promises mean, and that’s when the real governance test begins.

Related: Pentagon officials sent Anthropic best and final offer for military use of its AI amid dispute

Compliance that helps you close $1M+ deals

Comp AI helps companies get SOC 2, ISO 27001, HIPAA, or GDPR compliant in > 10 hours. It connects to your stack and keeps you audit-ready year-round, hands-off.

Teams like Dub, Strix, and Better Auth used it to unlock enterprise deals.

Superpower Daily readers get $2,000 off.

Jack Dorsey just halved the size of Block’s employee base — and he says your company is next

Key Takeaway: Block is cutting ~4,000 employees—nearly half its workforce—because Dorsey believes AI will let smaller teams move faster, and he predicts most companies will follow within a year.

More Insights:

• Block’s headcount drops from 10,000+ to just under 6,000 across Square, Cash App, and Tidal.

• Investors cheered the move, pushing the stock up ~24% after hours.

• Dorsey framed the cuts as “proactive” to avoid repeated layoffs that erode morale and trust.

• CFO Amrita Ahuja said AI automation will enable “smaller, highly talented teams” to do more work.

• Severance for U.S. employees includes 20 weeks pay + 1 week per year, equity vesting through end of May, 6 months health care, devices, and $5,000 transition support.

Why it matters: If leaders can successfully point to AI as the reason they need half the people, “efficiency” stops being a strategy and becomes a competitive requirement—meaning the real job security shift won’t be about performance alone, but about whether your work is legible to automation and scalable by smaller teams.

Uber engineers built an AI version of their boss

Key Takeaway: Uber CEO Dara Khosrowshahi says teams created an AI version of him to rehearse presentations, reflecting how deeply AI tools are reshaping engineering work and productivity at Uber.

More Insights:

• Khosrowshahi frames Uber as “a giant code base,” with engineers as the core builders of the company.

• Some teams reportedly use “Dara AI” to practice pitch decks before presenting to leadership.

• He claims ~90% of Uber’s software engineers use AI tools in their work.

• About 30% are “power users,” using AI to rethink and redesign parts of Uber’s architecture.

• Khosrowshahi says the productivity gains from AI are unlike anything he’s seen before.

Why it matters: AI isn’t just speeding up tasks—it’s changing how decisions get shaped before leaders ever enter the room; when teams can simulate executive reactions, the real shift is that influence and strategy start getting “trained” into the workflow itself.

Other stuff

• Burger King will use AI to check if employees say ‘please’ and ‘thank you’

• Most teens believe their peers are using AI to cheat in school

• Hacker used Anthropic's Claude chatbot to attack multiple government agencies in Mexico

• ‘Probably’ doesn’t mean the same thing to your AI as it does to you

• Michael Pollan Punctures the AI Bubble

• So, we’re getting Prada Meta AI glasses, right?

Hope you enjoyed today's newsletter

⚡️ Join over 300,000 people using the Superpower ChatGPT extension on Chrome and Firefox.