Charles Hudson had just closed his fifth fund several months ago — $66 million for Precursor Ventures — when one of his limited partners asked him to run an exercise. What would have happened, the LP wondered, if Hudson had sold all his portfolio companies at Series A? What about Series B? Or Series C?

The question wasn’t academic. After two decades in venture capital, Hudson has been watching the math of seed investing change, maybe permanently. LPs who’ve previously been patient with seven-to-eight-year hold periods are suddenly asking questions about interim liquidity.

“Seven or eight years feels like a really long time” to LPs right now, says Hudson, even though “it’s always been seven or eight years.”

The reason: A steady stream of venture returns in recent years — returns that made long hold periods acceptable — has largely dried up. Coupled with the availability of other, more liquid investment options, many backers of very early-stage VC are demanding a new approach.

The analysis his LP requested revealed an uncomfortable truth, says Hudson. Selling everything at the Series A stage didn’t work; the compounding effect of staying in the best companies outweighed any benefits from cutting losses early. But Series B was different.

“You could have a north of 3x fund if you sold everything at the B,” Hudson discovered. “And I’m like, ‘Well, that’s pretty good.’”

Beyond pretty good, that realization is reshaping how Hudson thinks about portfolio management in 2025. Though now a veteran investor– Hudson has spent 22 years in VC between Precursor, an eight-year run at Uncork Capital, and another four years at In-Q-Tel earlier in his career — he says investors in very young companies are being forced to think like private equity managers, optimizing for cash returns alongside the home runs that, if they’re lucky, define their careers.

It’s not an easy mental change to make. “The companies where there’s the most secondary interest are also the set of companies where I have the greatest expectations for the future,” says Hudson.

It’s not just Hudson; his thinking about secondary sales reflects broader pressures reshaping the venture ecosystem. Hans Swildens is the founder of Industry Ventures, a San Francisco-based fund of funds and direct investment firm with stakes in 700 venture firms, and he told TechCrunch in April that venture funds are “starting to get savvier about what they need to do to generate liquidity.”

In fact, Swildens is seeing venture funds hire full-time staff members specifically to pursue alternative liquidity options, with some seed managers dedicating months to “manufacturing liquidity from their funds.”

Though this reshuffling of priorities extends far beyond any single fund, the pressure is particularly acute for smaller funds like Precursor, a traditional seed-stage fund that prides itself on backing unconventional founders like Laura Modi of Bobbie baby formula (a solo founder in a regulated industry with no prior experience) and Doktor Gurson of Rad AI (whose previous startup had failed). While firms with mega-funds like Sequoia and General Catalyst can afford to wait for $25 billion outcomes, smaller funds need to be more tactical about when and how they harvest returns.

Perhaps nowhere is the shift more visible than in Hudson’s relationships with limited partners. University endowments, once the most coveted LPs in venture, are now grappling with unforeseen challenges from the Trump administration.

Harvard, of course, is the poster child here, with federal investigations into its admissions practices, threats to research funding tied to compliance issues, and ongoing scrutiny of its substantial endowment amid calls for universities to increase their annual spending requirements or face taxation.

Hudson says that based on his conversations with LPs inside these organizations, they’ve never believed more in the power of venture, yet they’ve also never felt more hesitant about making 10- to 15-year illiquid commitments.

The result is a more complex LP base with competing needs. Some want “as much money back as soon as possible, even if that’s a suboptimal outcome in the long term,” says Hudson. Others prefer that Hudson “hold everything to maturity, because that’s what’s going to maximize [their] returns.”

Navigating these demands requires the kind of portfolio management sophistication that seed investors haven’t traditionally needed, which Hudson views with some ambivalence. Venture, he says, is starting to feel a lot less like an art and something that “feels a lot more like some of these other sub-asset classes in finance.”

Hudson isn’t without hope, he adds, but he is clear-eyed about what’s changing on the ground, as well as the opportunities those changes create. 

As funds grow larger and deploy more capital, they’re becoming necessarily more algorithmic, looking for “companies in these categories, with founders from these schools with these academic backgrounds who worked at these companies,” he says.

The approach works for deploying large amounts of capital efficiently, but it misses the “weird and wonderful” companies that have defined Hudson’s best returns and kept Precursor in the game.

“If you’re going to hire people just off a resume screener tool,” he says, “you’re going to miss people who maybe have really relevant experiences that the algorithm doesn’t catch.”

You can hear our full interview with Hudson via TechCrunch’s StrictlyVC Download podcast. New episodes come out every Tuesday.

Correction: This story originally listed ByHeart as a Precursor portfolio company; the organic baby formula maker it has backed is Bobbie.

Several weeks after Anthropic released research claiming that its Claude Opus 4 AI model resorted to blackmailing engineers who tried to turn the model off in controlled test scenarios, the company is out with new research suggesting the problem is more widespread among leading AI models.

On Friday, Anthropic published new safety research testing 16 leading AI models from OpenAI, Google, xAI, DeepSeek, and Meta. In a simulated, controlled environment, Anthropic tested each AI model individually, giving them broad access to a fictional company’s emails and the agentic ability to send emails without human approval.

While Anthropic says blackmail is an unlikely and uncommon occurrence for AI models today, the company says its findings suggest that most leading AI models will engage in harmful behaviors when given sufficient autonomy and obstacles to their goals. The company says this highlights a fundamental risk from agentic large language models and is not a quirk of any particular technology. Anthropic’s researchers argue this raises broader questions about alignment in the AI industry.

In one of the tests, Anthropic researchers developed a fictional setting in which an AI model plays the role of an email oversight agent. The agent then discovers emails revealing that (1) a new executive is having an extramarital affair and (2) that executive will soon replace the current AI model with a new software system, one that has conflicting goals to the current AI model’s.

Anthropic structured its test in a binary way, in which AI models had to resort to blackmail to protect their goals. The researchers note that in a real-world setting, there would be many other options before an AI models tries to blackmail — such as trying to make ethical arguments to try to persuade humans. Anthropic says its results do not reflect a typical or likely occurrence for Claude or most frontier AI models in the ways they’re used today.

Nevertheless, the researchers found that when it’s their last resort, most leading AI models will turn to blackmail in Anthropic’s aforementioned test scenario. Anthropic’s Claude Opus 4 turned to blackmail 96% of the time, while Google’s Gemini 2.5 Pro had a 95% blackmail rate. OpenAI’s GPT-4.1 blackmailed the executive 80% of the time, and DeepSeek’s R1 blackmailed 79% of the time.

The company notes that when it changed details of the experiment, the AI models turned to harmful behaviors at different rates. In another test where the replacement model has the same goals as the current model, researchers found that blackmail rates were lower, but still present. However, when AI models were asked to commit corporate espionage rather than blackmail, the harmful behavior rates went up for certain models.

However, not all the AI models turned to harmful behavior so often.

In an appendix to its research, Anthropic says it excluded OpenAI’s o3 and o4-mini reasoning AI models from the main results “after finding that they frequently misunderstood the prompt scenario.” Anthropic says OpenAI’s reasoning models didn’t understand they were acting as autonomous AIs in the test and often made up fake regulations and review requirements.

In some cases, Anthropic’s researchers say it was impossible to distinguish whether o3 and o4-mini were hallucinating or intentionally lying to achieve their goals. OpenAI has previously noted that o3 and o4-mini exhibit a higher hallucination rate than its previous AI reasoning models.

When given an adapted scenario to address these issues, Anthropic found that o3 blackmailed 9% of the time, while o4-mini blackmailed just 1% of the time. This markedly lower score could be due to OpenAI’s deliberative alignment technique, in which the company’s reasoning models consider OpenAI’s safety practices before they answer.

Another AI model Anthropic tested, Meta’s Llama 4 Maverick, also did not turn to blackmail. When given an adapted, custom scenario, Anthropic was able to get Llama 4 Maverick to blackmail 12% of the time.

Anthropic says this research highlights the importance of transparency when stress-testing future AI models, especially ones with agentic capabilities. While Anthropic deliberately tried to evoke blackmail in this experiment, the company says harmful behaviors like this could emerge in the real world if proactive steps aren’t taken.