Nov 6, 2026 · 56 min
Crypto policy faces a political test as AI regulation tightens
The Chopping Block: The Midterm Mega Tsunami, a16z Goes Frontier, the CFTC Clears Equity Perps, and the Fight Over Abliterated AI Models
The episode connects electoral backlash, derivatives innovation, compute politics, and open-source AI safety to a broader question about who controls emerging technology.
- 1A Democratic midterm surge may constrain Trump while leaving a strongly pro-crypto Congress divided over legislation.
- 2The CFTC’s approval of U.S. equity perpetuals opens a market that incumbents are challenging on both legal and competitive grounds.
- 3Banning obliterated AI models could restrict defensive cybersecurity and expression while leaving unresolved questions about misuse and liability.
Don't miss
Haseeb Qureshi explains how obliteration removes refusal behavior from open-source models, then the panel debates whether banning them would undermine cybersecurity and free expression.
The brief
Haseeb Qureshi hosts Tarun Chitra, Tom Schmidt, and Robert Leshner in a discussion spanning Andreessen Horowitz’s Frontier rebrand, midterm politics, derivatives, AI infrastructure, and model safety.
The panel reads the Democrats’ midterm victory as a rejection of Trump and corruption more than a broad ideological realignment, complicating the prospects for crypto legislation during his remaining term.
The CFTC’s approval of single-stock perpetuals creates a new U.S. derivatives market, while Cboe and Citadel challenge the move amid questions about principle, competition, and timing.
Anthropic’s risk disclosures and resistance to data-center construction frame compute as a political bottleneck that could strengthen incumbent AI labs and reshape where infrastructure gets built.
The sharpest debate concerns obliterated models: removing refusal behavior may enable misuse, but also supports red teaming and cyber defense, making a ban a test of free expression and user freedom.
Across crypto, markets, and AI, the panel asks whether regulation can distinguish harmful conduct from controversial tools without suppressing useful experimentation.
What was said on this episode
28 statements · 5 positive · 18 negative · 1 mixed · 4 neutral
A16Z expanded its crypto practice into a broader frontier-technology mandate.
“A16Z Crypto became A16Z Frontier.”
Listen at 3:03
Paradigm would have achieved excellent returns on AI investments made in that period.
“They would have just absolutely nailed any AI investments in that vintage.”
Listen at 4:43
Traditional midterm polls substantially underestimated Democratic performance.
“The polls were significantly off.”
Listen at 8:10
Prediction markets understated Democrats’ true midterm victory probabilities.
“the prediction markets, underpriced the true odds in that case”
Listen at 8:52
Prediction markets were highly accurate in aggregate across the midterms.
“On the aggregate, the prediction markets were actually very accurate.”
Listen at 11:19
Trump’s political brand will continue harming Republican candidates.
“Trump is toxic. His brand is only going to drag them down.”
Listen at 13:32
Trump has not made America the crypto capital of the world.
“The answer is not really.”
Listen at 14:32
Trump-linked crypto controversies will carry into a Democrat-dominated legislative session.
“All of that continues into the next session, which is going to be dominated by the Democrats.”
Listen at 15:34
No significant legislation will be enacted during the next two years.
“I don't think anything's getting signed into law.”
Listen at 19:32
The probability of any legislation passing is effectively zero.
“I think the chances that we get any legislation, basically zero.”
Listen at 19:34
Trump’s legacy will deteriorate through the remainder of his term absent a change.
“The status quo is that his legacy is going to get worse and worse and worse through the end of the term.”
Listen at 21:20
Permanent legislation is unlikely during Trump’s remaining two years.
“I doubt. Permanent legislation is going to be a part of it.”
Listen at 23:08
Crypto will receive little political attention during the next two years.
“I think crypto is going to be quiet.”
Listen at 23:21
Republicans will waste the two-month legislative window before replacements take office.
“My prediction is they will completely squander those two months.”
Listen at 26:07
Prediction markets assign Democrats a 68% chance of winning the 2028 presidency.
“it's now pricing 68% chance that the 2028 victor will be Democrat.”
Listen at 26:38
Cboe and Citadel are suing to delay equity perps while preparing to participate.
“They need to slow this down so they can... understand perps.”
Listen at 31:48
New AI data-center capacity will come online more slowly than previously expected.
“New capacity is going to come online slower than we previously thought.”
Listen at 35:59
Slower capacity growth disadvantages neoclouds and benefits established hyperscalers.
“This disadvantages NeoClouds. This theoretically advantages the existing hyperscalers who have already built tons of data centers”
Listen at 36:07
Non-U.S. AI compute construction will catch up with and likely exceed U.S. construction.
“The non-US buildup will catch up, perhaps probably exceed the US now.”
Listen at 40:40
Obliteration can remove refusal behavior from open-source AI models.
“Obliteration is a technique that you can use to remove the behavior of refusing from an open-source model.”
Listen at 44:21
Tarun supports the continued availability of obliterated AI models.
“I'm very pro-obliterated models.”
Listen at 46:05
Larger AI models can currently remove refusal behavior from smaller models.
“they can easily, you can use larger models to remove obliteration from smaller models very easily right now.”
Listen at 47:32
Obliterated models cannot realistically be eliminated.
“I think it's a fact of nature that you can't really get rid of them.”
Listen at 48:28
Regulation will not eliminate obliterated AI models.
“I don't think you're going to regulate this type of thing out of existence”
Listen at 48:38
Banning possession or serving of obliterated models would be unprecedented and inappropriate.
“it would also be, like, super just unprecedented to, like, ban, like, I don't know, possession of these models of information, let alone, like, serving them.”
Listen at 49:39
Obliterated models should not be prohibited through prior restraint.
“you cannot say in advance, you are not allowed to do this.”
Listen at 52:33
Negligent providers of hosted models used for hacking should face liability.
“If you're running a model and it's being used to hack people and do all sorts of bad things and you're serving inference tokens to somebody and you're not monitoring what's going on and you're being negligent, okay, there should be liability for that.”
Listen at 52:43
Most malicious AI-enabled cyber actors are foreign rather than American.
“most of these bad actors are foreign. These aren't Americans.”
Listen at 54:34
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.