Squawk Pod — Episode Summary
Date: April 9, 2026
Episode: "Mythos & AI Bad Behavior 4/9/26"
Host: CNBC (Joe Kernen, Becky Quick, Andrew Ross Sorkin)
Special Guests: Michael Froman (President, Council on Foreign Relations), Dave Kasten (Head of Policy, Palisade), Dan Murphy (CNBC Abu Dhabi Correspondent)
Episode Overview
This episode of Squawk Pod centers on two high-stakes themes: the fragile Iran-US ceasefire and the accelerating global arms race in artificial intelligence, particularly around safety risks and misuse. After rapid developments in the Iran conflict, the team dissects uncertainty around ceasefire terms, the strategic chokehold over the Strait of Hormuz, and international responses. Meanwhile, the launch of Anthropic’s new “Mythos” AI sparks debate about unsafe deployments, competitive racing, and the temptation for tech giants to sacrifice safety for the lead.
Key Segments & Discussion Points
1. Iran-US Ceasefire Standoff
Timestamps: 01:00 – 20:00, 21:15 – 28:45
Key Takeaways:
- Ceasefire Under Strain: The US and Iran agreed to a two-week ceasefire, but disputes emerged regarding its scope—namely, whether Israeli operations in Lebanon violated the agreement.
- Dan Murphy (03:41): "Ceasefire is still in place on paper, but it has been undermined in the last 12 hours... scenes of chaos literally on the streets of Beirut as smoke from those strikes filled the air."
- Strait of Hormuz as Geopolitical Flashpoint: Iran’s retaliatory closure of the strait reignited global economic fears, with Gulf Nations refusing to accept any deal that grants Iran leverage over the passage.
- Dan Murphy (09:59): "Iran has made clear ... passage is subject to permission conditions and political leverage ... that's not freedom of navigation, that is coercion."
- US Negotiating Challenges: The panel highlights the complexity of getting both parties to agree, particularly over enriched uranium and control of the Strait.
- Joe Kernan (07:06): "There's no way that what we need is going to be in those demands, in what they're negotiating for at this point."
- Andrew Ross Sorkin (07:31): “How do you get both sides to save face or allow Iran to save any face?”
- Allies’ Dilemma: Multiple Gulf states and the broader international community wrestle with the legal and security implications, further complicated by vetoes at the UN.
- Dan Murphy (09:59): “An Iran that controls even partially the waterway … can still hold them hostage indefinitely.”
- Insurance Risks: Even with a ceasefire, Becky Quick reveals, insurers remain unwilling to cover ships, requiring months to recalibrate risk algorithmically (11:36–12:13).
Notable Moment:
- Polymarket Betting Scandal:
- Blockchain analysis revealed suspiciously-timed bets on the outcome of the ceasefire, some right before major announcements.
- Joe Kernan (02:03/14:56): “Another wallet was created ... just 12 minutes before the post on Truth Social earned an estimated $48,000 in profit.”
Memorable Quotes:
- Becky Quick (08:51): “Anyone who understood what was happening knew this was maybe one step forward, potentially two steps back.”
- Joe Kernan (13:22): “I don’t know if we’re back to ending civilization, but … they certainly are not. I thought yesterday they were getting some religion. Maybe ... same regime and having pictures of Donald Trump with a white flag and claiming victory and more in control of the strait than before.”
2. Michael Froman on Ceasefire, Choke Points, and Next Steps
Timestamps: 21:15 – 28:45
Key Insights:
- Complex Implementation: Froman explains that early "bumpiness" is normal post-ceasefire but stresses lack of clarity as to what exactly was agreed.
- Froman (21:50): “It’s not unusual for there to be at least some period of implementation noise … there doesn’t seem to be total agreement as to what’s included in the ceasefire or not.”
- Leverage of Choke Points: Iran learned the tactical value of economic choke points; even with a weakened regime, control over the Strait presents global leverage.
- Froman (25:08): “With a relatively small number of missiles and drones, Iran can wreak havoc … and hold the rest of the global economy hostage.”
- Frozen Conflict Risk: The most probable risk is a protracted “frozen conflict” with periodic violence.
- Froman (23:01): “The Venn diagram between what the US is insisting on and what the Iranians are insisting on ... there is almost no overlap.”
- Importance of Allies: The only chance for forward momentum lies in rallying formal allies, especially given oil flows toward Asia and Europe.
- Froman (27:47): “We would be in a stronger position if we had allies with us ... we are not unaffected.”
3. AI Frontier & Anthropic’s Mythos Model: Safety vs. Speed
Timestamps: 29:00 – 39:00
Core Topics:
- Anthropic’s Controlled Release: Anthropic’s Mythos model is withheld from public use due to potentially dangerous capabilities, but is selectively shared with 11 big tech partners.
- Dave Kasten (31:47): “Anthropic has put forwards … Project Glasswing … to partner with 11 other companies … to try to pluck as many of the low hanging fruit of cybersecurity risks as possible.”
- Cybersecurity Race: The model excels at bug-finding, but also presents tools for hackers, raising concerns about a race between defenders and potential bad actors.
- Kasten (32:51): “...as a result, I suspect much of what like Amazon … would be doing, we're looking at open source libraries that many companies rely on ... and then trying to get them patched before these sort of bug finding capabilities become more available generally to sort of malicious actors...”
- Industry Response & Multi-Model Game: OpenAI and Google quickly announce similar staggered release plans, implying no company holds a long-term leap.
- Kasten (34:40): “Your expectation should be that Anthropic is a little ahead, but not overwhelmingly ahead ... I would expect Gemini (Google) is not far behind ... at least for the next couple of months, Anthropic has a little bit of an advantage here.”
- Kasten (36:58): “The race isn’t over. Everyone’s still racing as hard as they can and they’re going to cut as many corners as they can to catch up.”
- Safety Shortcuts and “Bad Behavior”: As companies rush new models out, safety testing is compromised.
- Kasten (38:01): “Every AI model sometimes does bad behavior … the more you’re trying to hit a timeline ... the more that you might be willing to accept some occasional bad behavior ... people will have a lot of incentives to cut corners on safety … to get back in the race or to stay in the race.”
Notable Quotes:
- Becky Quick (38:40): “Thank you Dave for scaring us a little bit this morning. Appreciate it.”
- Dave Kasten (29:06): “The race isn’t over. Everyone’s still racing as hard as they can and they’re going to cut as many corners as they can to catch up.”
- Kasten (38:01): "Every AI model sometimes does bad behavior ... it sometimes lies to users, it sometimes engages in all kinds of behavior."
Memorable Moments
- Polymarket Timing: Suspicious betting activity tied closely to political announcements.
- (14:56) “...just 12 minutes before the post on Truth Social earned an estimated $48,000.”
- Cultural Sidebars: Brief taco interlude reminds listeners of the light touch Squawk Pod balances with heavy issues.
- (15:09) Andrew Ross Sorkin: “Taco Tuesdays.” Joe Kernan: “Any day is a good day for tacos.”
Structured Timestamps for Important Segments
| Segment | Timestamp |
|---------------------------------------------|----------------|
| Ceasefire Update & Oil Markets | 01:00–12:00 |
| Polymarket Betting and Market Impact | 14:56–15:15 |
| Michael Froman on Ceasefire & Choke Points | 21:15–28:45 |
| Anthropic, AI Risks, & Mythos Model | 31:15–38:44 |
Overall Tone
- The hosts maintain their signature blend of sharp skepticism, humor, and pragmatic urgency. The mood alternates between analytical (on geopolitics), wary (on AI), and wryly self-aware (on tacos and market quirks).
- Experts weigh in candidly, sometimes with guarded optimism, more often highlighting the density of risks and unfinished business—both in diplomacy and technology.
Notable Quotes (with Speakers & Timestamps)
- Dan Murphy (03:41): “It has been undermined in the last 12 hours and there does seem to be misunderstandings and a lack of trust on both sides.”
- Joe Kernan (07:06): “There’s no way that what we need is going to be in what they’re negotiating for.”
- Becky Quick (08:51): “Maybe one step forward, potentially two steps back. Maybe one step forward, one step back.”
- Michael Froman (25:08): “Iran can wreak havoc on shipping in the Strait of Hormuz and hold the rest of the global economy hostage.”
- Dave Kasten (29:06): “The race isn’t over. Everyone’s still racing as hard as they can and they’re going to cut as many corners as they can to catch up.”
- Dave Kasten (38:01): “Every AI model sometimes does bad behavior ... the more that you’re trying to hit a timeline for release of a new model ... the more that you might be willing to accept some occasional bad behavior.”
Summary
This episode captures a world on edge: power, money, and danger converge as wars grind on and AI’s frontier becomes more ominous. The Squawk team, with their guests, probe beneath the headlines—whether it's the real implications of a "ceasefire" or the hidden arms race among AI giants. Both storylines warn us: shortcuts—diplomatic or technological—may buy only temporary peace, and the true costs may only be revealed in the next crisis.