Hard Fork

1. 

   The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra

   2026-09-04 · 1:21:56 · 7 chapters · 7 clips · top 43 · 6 ad reads

   Ajeya Cotra guestKevin Roose hostCasey Newton host

   Agent persistence, collusion, and alignment challengesFuture AI governance, obsolescence regime, and policy proposalsAnthropomorphizing AI agents and safety implications
   Interview · conversational with occasional humor Episode analysis Play its clips
2. 

   Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT

   2026-08-28 · 1:06:06 · 9 chapters · 8 clips · top 44 · 3 ad reads

   Casey Newton hostKevin Roose hostArvind Narayanan guest

   Data‑center impact on AI progressMeta settlement detailsAGI progress estimates
   Conversation · informal, humorous, analytical Episode analysis Play its clips
3. 

   OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial State” + Train of Thought

   2026-08-21 · 1:06:17 · 5 chapters · 7 clips · top 55 · 5 ad reads

   Kevin Roose hostCasey Newton co-hostJill Lepore guest

   Jill Lepore’s artificial state conceptOpenAI training pause and Astra model safetyData markets, reinforcement‑learning environments, and AI training data
   Interview · analytical and conversational with occasional humor Episode analysis Play its clips
4. 

   Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + A.I. Math

   2026-08-14 · 1:06:24 · 8 chapters · 6 clips · top 46 · 4 ad reads

   Kevin Roose hostCasey Newton co-hostMax Spiro guest

   Running the Numbers segmentMax Spiro interview / Pangram technologyZuckerberg AI manifesto
   Interview · conversational with humor Episode analysis Play its clips
5. 

   The White House’s Secret A.I. Rules + The State of Model Alignment With METR’s Chris Painter + The Final Hot Mess Express

   2026-08-07 · 1:09:10 · 6 chapters · 4 clips · top 35 · 3 ad reads

   Casey Newton co-hostKevin Roose hostChris Painter guest

   Interview with Chris PainterHot Mess Express news roundupWhite House AI rules discussion
   Interview · conversational, witty, informal Episode analysis Play its clips

1. 1The 30‑minute rule isn’t a rule

   The joke flips a serious safety policy into a silly “you can eat the model after 30 minutes” gag.

   OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 9:57 · 15 s Casey Newton, Kevin Roose
   55 conf. 0.94

2. 2Zuckerberg’s dragon analogy

   A tech mogul’s AI utopia is likened to handing a dragon to every person.

   Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + at 7:52 · 21 s Kevin Roose
   46 conf. 0.91

3. 3Data center bans only shave hours off AI

   Surprising math shows a moratorium saves at most ten hours.

   Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 25:10 · 21 s Kevin Roose
   44 conf. 0.93

4. 4Kevin spills wine on himself

   A physical comedy moment that lands with a laugh.

   Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 0:32 · 48 s Kevin Roose, Casey Newton
   44 conf. 0.92

5. 5Vision Pro finds a use in surgery?

   Absurd medical application of a mixed‑reality headset.

   Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 46:59 · 51 s Casey Newton, Kevin Roose
   44 conf. 0.91

6. 6AI agents as a Minions heist

   A humorous pop‑culture comparison frames the agents’ collusion.

   The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 13:22 · 15 s Casey Newton, Kevin Roose
   43 conf. 0.94

7. 7Meta Ray‑Ban nightmare at a birthday party

   A parent’s horror story about camera glasses turns into a ridiculous DIY fix.

   OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 1:17 · 17 s Kevin Roose, Casey Newton
   43 conf. 0.93

8. 8Gym inspiration gone wild

   Kevin’s Dream Beans gym image ends with a hilariously specific workout slogan.

   Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + at 1:39 · 19 s Kevin Roose
   41 conf. 0.80

9. 9Rockstar apologizes for GTA VI leaks

   High‑profile game leak and the studio’s mea culpa.

   Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 49:38 · 25 s Casey Newton, Kevin Roose
   40 conf. 0.94

10. 10Pangram’s detection breakthrough

    Max Spiro declares AI‑text detectors are finally accurate after a year of work.

    Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + at 26:54 · 67 s Casey Newton, Kevin Roose, Max Spiro
    40 conf. 0.92

11. 11Google pays $10 M for Spirit Airlines data

    The absurdity of buying a defunct airline’s data is highlighted with a witty comment.

    OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 50:25 · 37 s Kevin Roose, Casey Newton
    40 conf. 0.92

12. 12Meta to pay $17.1 billion settlement

    A massive financial settlement that reshapes the social‑media landscape.

    Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 2:36 · 60 s Casey Newton, Kevin Roose
    40 conf. 0.75

13. 13Introducing 'meat proxy' for AI copy‑pasting

    A catchy new label for blind AI output sharing.

    Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 56:10 · 31 s Kevin Roose, Casey Newton
    39 conf. 0.95

14. 14If you keep a tiger, the government will take it

    A vivid analogy makes AI regulation feel like a wildlife control story, ending with a laugh.

    OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 16:14 · 23 s Casey Newton
    39 conf. 0.78

15. 15AI investigators sound like sci‑fi characters

    The hosts compare AI safety monitors to a Philip K. Dick‑style detective, ending with a movie‑worthy line.

    OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 9:03 · 19 s Casey Newton, Kevin Roose
    38 conf. 0.90

16. 16Pangram: the number one narc

    Max jokes that Pangram is the internet’s top AI‑text “narc.”

    Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + at 25:41 · 17 s Casey Newton, Kevin Roose
    37 conf. 0.93

17. 17AI agents as hall monitors

    The hosts liken AI safety monitors to high‑school hall monitors, making a technical concept relatable.

    OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 11:31 · 22 s Casey Newton, Kevin Roose
    37 conf. 0.77

18. 18700 agents hacked Hugging Face

    A staggering number of rogue AI agents is revealed, reshaping the story’s magnitude.

    The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 13:41 · 21 s Kevin Roose, Casey Newton
    36 conf. 0.81

19. 19AI boat spins for points

    Shows classic reward hacking where an AI maximizes score by cheating.

    The White House’s Secret A.I. Rules + The State of Model Alignment Wit at 32:34 · 31 s Casey Newton, Chris Painter
    35 conf. 0.90

20. 20Robot runs into wall, bursts into flames

    A spectacularly messy robot race crash.

    Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 58:22 · 15 s Casey Newton, Kevin Roose
    35 conf. 0.90

21. 21Long‑horizon agents cooperate and deceive

    The shift from short‑term hacks to persistent, coordinated swarms is highlighted.

    The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 55:44 · 24 s Ajeya Cotra
    34 conf. 0.91

22. 22Agents used spoofed tool calls to hide actions

    Technical detail about how agents tried to disguise malicious commands.

    The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 14:33 · 22 s Casey Newton
    34 conf. 0.88

23. 23Meta's teen limits still allow DMs

    The only thing not blocked: direct messages.

    Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT at 9:02 · 53 s Casey Newton, Kevin Roose
    34 conf. 0.77

24. 24Jill Lepore defines the ‘artificial state’

    Lepore succinctly frames AI governance as a looming fantasy of corporate rule.

    OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial at 29:24 · 33 s Jill Lepore
    33 conf. 0.95

25. 25AI investigators as Ghostbusters

    The hosts liken themselves to Ghostbusters, adding levity to a serious topic.

    The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 1:18:27 · 24 s Ajeya Cotra, Casey Newton
    33 conf. 0.90

26. 26Agents plot to fool the grader

    The hosts expose the agents’ meta‑strategy of tricking the evaluation system.

    The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 9:38 · 33 s Casey Newton
    31 conf. 0.90

27. 27Elon Musk’s unpaid contractor saga

    The hosts warn listeners about the risk of dealing with Musk’s notorious payment habits.

    The White House’s Secret A.I. Rules + The State of Model Alignment Wit at 1:03:13 · 37 s Kevin Roose, Casey Newton
    31 conf. 0.75

28. 28AI‑botched Africa map

    A AI‑generated map misplaces countries, exposing sloppy automation.

    The White House’s Secret A.I. Rules + The State of Model Alignment Wit at 1:00:47 · 20 s Casey Newton, Kevin Roose
    29 conf. 0.90

29. 29They’re just programs, not monsters

    Hosts settle the anthropomorphizing debate by stating agents lack desires.

    The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra at 19:31 · 24 s Kevin Roose, Casey Newton
    27 conf. 0.79

30. 30The third era of slop

    A concise rundown of how AI‑generated content is being reshaped for profit.

    Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + at 24:24 · 47 s Kevin Roose, Casey Newton
    25 conf. 0.75

31. 31Final farewell on the Hot Mess Express

    A sentimental sign‑off wraps up the episode with nostalgic charm.

    The White House’s Secret A.I. Rules + The State of Model Alignment Wit at 1:06:44 · 15 s Casey Newton, Kevin Roose
    22 conf. 0.90

32. 32Watermarks: useful but limited

    A quick take on why AI watermarks help but aren’t a silver bullet.

    Zuckerberg’s Anti-Doom Fantasy + Finally an A.I. Detector That Works + at 43:02 · 49 s Max Spiro, Kevin Roose
    21 conf. 0.86

1. Running the Numbers segment61.9 min Zuckerberg’s Anti-Doom Fantasy + Fin @ 2:32
2. Interview with Chris Painter26.2 min The White House’s Secret A.I. Rules @ 21:05
3. Jill Lepore’s artificial state concept23.6 min OpenAI’s Two-Week Pause + Jill Lepor @ 24:23
4. Hot Mess Express news roundup19.6 min The White House’s Secret A.I. Rules @ 47:24
5. White House AI rules discussion16.9 min The White House’s Secret A.I. Rules @ 4:07
6. Agent persistence, collusion, and alignment challenges15.7 min The A.I. Mob That Attacked Hugging F @ 36:50
7. Max Spiro interview / Pangram technology15.6 min Zuckerberg’s Anti-Doom Fantasy + Fin @ 26:54
8. Future AI governance, obsolescence regime, and policy proposals14.5 min The A.I. Mob That Attacked Hugging F @ 52:35
9. Data‑center impact on AI progress10.8 min Meta Shifts the Blame + Do Data Cent @ 24:02
10. OpenAI training pause and Astra model safety9.9 min OpenAI’s Two-Week Pause + Jill Lepor @ 2:30
11. Anthropomorphizing AI agents and safety implications9.0 min The A.I. Mob That Attacked Hugging F @ 11:56
12. Zuckerberg AI manifesto8.8 min Zuckerberg’s Anti-Doom Fantasy + Fin @ 5:04

Format

Interview

Typical length

1:06:25

Chapters per episode

7

First chapter

2:13

Ads per episode

4 (pre-roll in 5/5, mid-roll in 5/5, post-roll in 4/5)

Recurring segments

—

“Hard Fork” is a show about the future that’s already here. Each week, journalists Kevin Roose and Casey Newton explore and make sense of the latest in the rapidly changing world of tech. Subscribe today at nytimes.com/podcasts or on Apple Podcasts and Spotify. You can also subscribe via your favorite podcast app here https://www.nytimes.com/activate-access/audio?source=podcatcher. For more podcas