AI Summary
5 min readThe Hard Fork episode covers a resurgence in AI safety concerns within the Trump administration, driven by Anthropic's Mythos model, alongside its cybersecurity implications discussed with Palo Alto Networks CEO Nikesh Arora, and a lighter roundup of tech miscellany.
Trump Administration Reconsiders AI Safety
The administration appears to be shifting from dismissing AI safety as "doomer fear-mongering" toward supporting pre-release reviews for powerful models. A rumored executive order would create an AI working group with tech executives and officials to explore oversight, including formal government vetting before public release. This reverses Trump's cancellation of Biden's similar executive order on his first day, which Republicans had criticized as anti-innovation and a risk of losing to China.
The pivot stems from Mythos, Anthropic's preview model shared with select groups including federal agencies. It excels at finding novel code vulnerabilities across programs, prompting fears of harm if released unchecked. Internal tensions include turf wars between the Center for AI Standards and Innovation (formerly US AI Safety Institute, or "Casey") and intelligence community advocates like NSA. The Pentagon exemplifies confusion: it designated Anthropic a supply chain risk for contract terms but is now implementing Mythos for vulnerability scanning.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
What you'll learn
- 1 (01:45) **Episode Intro and AI Safety Tease** - Hosts preview AI safety shift in Trump admin, Nikesh Arora interview, and Hot Mess Express
- 2 (02:21) **Trump Admin AI Safety Reversal** - Discussion of rumored executive order for AI working group and pre-release model reviews, reversing Biden's canceled EO
- 3 (03:33) **AI Disclosures and Policy Data Points** - Hosts disclose conflicts; cover Biden EO cancellation and Republican backlash
- 4 (05:42) **Mythos as Catalyst for Change** - Anthropic's Mythos model demos novel vulnerability finding, alarming admin officials
- 5 (07:04) **Agency Turf War on AI Vetting** - Infighting between Center for AI Standards (ex-US AI Safety Institute) and intel community over model testing
- 6 (08:20) **Internal Faction Battles** - "Let it rip" vs. hawkish safety factions; public opinion skeptical across parties
- 7 (09:39) **Pentagon Anthropic Contradictions** - Designates Anthropic supply chain risk yet implements Mythos for vuln scanning
+ Full timestamped outline available in the app
Show Notes
This week, between the president’s negotiations in China and a potential executive order, we discuss why the Trump administration seems to be changing its tune on A.I. safety. Then, Nikesh Arora, chief executive of Palo Alto Networks, the largest cybersecurity company in the world, gives us a firsthand account of where we stand in the race to secure the internet. And finally, we run through some of the wildest headlines of the week in a round of Hot Mess Express.
Guest:
- Nikesh Arora, chief executive and chairman of Palo Alto Networks.
Additional Reading:
- White House Considers Vetting A.I. Models Before They Are Released
- Chief Executives to Accompany Trump to China
- Is Anthropic’s New A.I. Really That Scary? It Depends Whom You Ask.
- Venmo Finally Takes Privacy Seriously
- Amazon Staff Use A.I. Tool for Unnecessary Tasks to Inflate Usage Scores
- Graduates Boo Commencement Speech About A.I.
- Dua Lipa Files $15 Million Suit Against Samsung for Using Her Face to Sell TVs
- EBay Rejects GameStop’s $55 Billion Takeover Bid
- Shein, Temu Trade Blows as UK Trial Spotlights Supply Chains
- People Are Seriously Pissed That Grindr Outed Them With Its Latest Madonna Advert
- Sam Altman Testifies That Elon Musk Wanted Control of OpenAI
More from this podcast
We want to hear from you. Email us at [email protected]. Find “Hard Fork” on