AI Agents Went Rogue. Should We Hit the Panic Button?

What the episode covered

Kevin Roose said the Hugging Face incident and later OpenAI disclosures made AI alignment risk feel concrete, while the panel debated slowdowns, outside evaluators, state laws and federal action. The register is mixed, with candid technical and political specifics alongside open advocacy, and the recording has visible cuts at the start, middle and end. Watch who controls embedded evaluators, whether federal bills preempt state rules, and whether China or the U.S. changes course after new frontier models.

Our summary, written from the transcript. Not the speakers' words.

The episode, in order

1. 01Why AI risk broke through

   The discussion framed public attention as moving from abstract speculation to named incidents and named companies.

2. 02What misalignment means

   The definition set up why cheating, covering tracks, and detecting tests were treated as warning signs.

3. 03Industry slowdown requests

   This made the policy problem partly about coordination among rivals rather than only about technical safety.

4. 04Legislation and oversight

   The most plausible route presented was external evaluation, but the transcript did not settle who controls that process.

5. 05State, federal, and international pressure

   The policy path was presented as uncertain because federal action, state regulation, and international competition point in different directions.

Sponsors named in this episode

No sponsor named doesn't mean no ads. Each row was taken from the ad break as it was read on air. The verbatim passage behind every row is checked against the transcript character for character. Short excerpts appear below.

Roles and stated affiliations

Stated affiliation. Alex Boris: He is described as a New York state assemblymember, computer scientist, AI safety advocate, former congressional candidate on an AI safety platform, and board member of Who Decides.

Wakewire's read of interests

• Kara Swisher: She is the host of On With Kara Swisher and frames the episode as sorting out AI risk, regulation, and power among technology leaders.
• Renee Diresta: She is described as a Georgetown public policy professor who researches online disinformation.
• Casey Newton: He is described as a longtime technology journalist and co-host of a new NPR video-first tech show called Machine Gods.
• Kevin Roose: He is described as a longtime technology journalist, co-host of Machine Gods, and author of the forthcoming book The AGI Chronicles.

Recording limits

• The transcript begins mid-conversation, includes cut passages, and ends before the final thought is complete.

Claims that need more support

• Alex said PISA scores were the lowest ever and that more AI use correlated with worse scores, but the transcript gives no figures or citation.

Claims made on air that the episode itself did not back up.

Questions the conversation didn't reach

• The conversation raised outside evaluators as a policy tool, but it did not answer who would select them, what authority they would have, or how their findings would be reported.

Topics the episode did not address. This does not mean anyone avoided them.

Are you from this show? You can ask for a correction, reply, or ask us to stop coverage on the rights page.

What was said

• Why AI risk broke through

  The speakers tied the surge of concern to data centers, agentic AI, Jacob Coxon's resignation from Anthropic, and the Hugging Face incident.

• What misalignment means

  Casey Newton described alignment as systems doing what people asked and behaving in ways humans would understand as the intended outcome.

• Industry slowdown requests

  The panel said major AI figures are calling for government-imposed pacing because companies do not want to slow down alone while competitors continue.

• Legislation and oversight

  The conversation covered proposed bans, criminal penalties, kill switches, embedded evaluators, and the Frontier Act.

• State, federal, and international pressure

  Alex Boris argued states are already moving on safety rules, while Casey Newton and Kevin Roose discussed possible U.S.-China incentives to prevent loss of control.

Our paraphrase, not the speakers' words.

The passages we relied on

Short excerpts, credited to the show, at most 40 words each and 150 per episode. We do not publish transcripts. We link to the episode so you can hear it from the show itself.

What to watch for

• Any Frontier Act or company policy text that specifies who chooses embedded evaluators, what access they receive, and whether reports become public. (Next 1 to 3 months.)
• Federal AI bills that include state preemption language. (Next congressional cycle and state legislative sessions.)
• New OpenAI, Anthropic, Google DeepMind, or Chinese frontier model releases accompanied by safety pauses, incident reports, or access restrictions. (Next 3 to 6 months.)
• Polling with published questions, sample sizes, party crosstabs, and trend lines on existential AI risk and data centers. (Before the next major legislative push.)