No rule or regulator required Anthropic to pause the release of its new Mythos model.
The San Francisco-based AI company voluntarily limited its distribution after finding that Mythos is able to identify thousands of previously unknown flaws in major operating systems and browsers – and exploit them within hours to gain full control of corporate networks.
As AI pioneer Yoshua Bengio is credited with observing, there’s more regulation of a sandwich in New York City than there is of the emergent power of artificial intelligence.
Mike Kubzansky, the outgoing head of Omidyar Network, urged impact investors to lean in, to invest up and down the AI technology “stack” – and to step up to the multiplying policy debates that will shape the future of technology and society.
“There’s a professional industry that wants you to think it’s too complicated and leave it to them,” Kubzansky told attendees of last week’s West Coast Impact Forum of the Global Impact Investing Network. “I’m here to say that’s bullshit and you should not be flummoxed by it.”
Kubzansky said his next act is the creation of an investor network, and affiliated fund, to help impact investors engage both with the builders of the technology and the policymakers that can shape it. As ImpactAlpha has reported, the vehicle, which had been code-named Project Alexandria, will invest in AI assurance and safety tech, such as algorithm auditing and deepfake detection, and call for sensible regulatory guardrails.
Kubzansky will discuss investor agency in the age of AI with Roy Swan of the Ford Foundation, Daryn Dodson of Illumen Capital and Bulbul Gupta of Pacific Community Ventures, in a conversation moderated by ImpactAlpha’s Dennis Price, at Mission Investors Exchange National Conference in Atlanta later this month.
“This is not a conventional impact story in the way we’re all used to thinking about it,” Kubzansky said at the GIIN forum. “But it’s equally important, in terms of driving a trustworthy stack that every citizen can feel confident using, that is looking out for their interests and not being extractive of them.”
Financial guardrails
The AI policy discussion heated up last week as well with OpenAI’s release of a white paper, “Industrial policy for the Intelligence Age.” The 13-page document articulated planks that some AI experts have proposed , including mechanisms to share the gains from AI-driven growth more widely to preserve what it called “shared prosperity.”
San Francisco-based OpenAI called for giving workers a voice in the AI transition, including formal collaboration with management to improve job quality, enhance safety and respect labor rights. It even suggested modernizing the tax base with higher taxes on capital gains, corporate income and taxes on automated labor.
But critics pointed out that, in practice, OpenAI has opposed regulations that could have advanced policies the company has supported in theory. OpenAI worked to weaken the European Union’s AI Act, which sought oversight of AI companies creating risky system. It also opposed the California’s Senate Bill 1047 which included risk-management strategies that OpenAI CEO Sam Altman had earlier proposed. OpenAI called for California Gov. Gavin Newsom to veto the legislation, as he eventually did.
“Perhaps, in some way, the paper extends a pattern described in a New Yorker profile of OpenAI founder and CEO Sam Altman, in which public pronouncements are paired with private contradictions,” Eryk Salvaggio wrote in Tech Policy.
A separate paper from the Vanderbilt Policy Accelerator at Vanderbilt University argued for financial guardrails in advance of what the center’s Asad Ramzanali argues is a coming crash sparked by overbuilding of AI data centers. He cites JPMorgan estimates of $5 trillion of AI infrastructure investment in the next five years, which would require annual revenues of $650 billion, several orders of magnitude higher than leading companies are generating.
“There’s massive disruption coming into the rest of your portfolios as well, which also needs to get navigated, which you’re going to need to be paying attention to,” Kubzansky said at the GIIN forum.
LP/GP
Kubzansky pointed to opportunities to invest across the technology stack of AI architecture, including the data center infrastructure and computing models and the “orchestration” layers that coordinate agents and other functions, as well as in “AI for good” applications in healthcare, education and scientific research.
Illumen Capital is seeking to ensure the outputs of AI system can be trusted to be bias-free. The fund of funds deploys bias-interruption modules to train AI builders, including its fund managers and their portfolio companies. Bias triggers are easy to spot, says Illumen’s Daryn Dodson. The goal is to interrupt those patterns so investors can see real market value that can be destroyed by race or gender bias.
At Pacific Community Ventures, Bulbul Gupta is using AI to advance economic mobility, including by training algorithms on inclusive investment data and community voices to speed lending to overlooked entrepreneurs. The community development financial institution last year acquired Radiant Data to bring AI to mission-based lending.
And there are plenty of startups pursuing valuable technologies to safeguard users and the broader public, but so far, there have been few dedicated investment vehicles focused on “good AI.”
The nonprofit Mozilla Foundation, an early thinker on building trustworthy AI systems, three years ago stood up Mozilla Ventures, a $35 million venture capital fund to invest in startups developing safe and inclusive AI. It has made some 55 investments to date, including in “trustworthy AI” companies such as Adaption Labs, Credo AI and Fiddler AI.
Mozilla Foundation’s Mark Surman has said the organization was forming “a rebel alliance of sorts,” to keep AI development more open and accountable in the face of the concentration of power in the tech behemoths.
Other funds, such as Juniper Ventures, are responding to demands to make AI secure and beneficial for humanity.
Omidyar Network, along with Ford Foundation and Nathan Cumings Foundation, two years ago took a small stake in Anthropic, which is planning an initial public offering, possibly later this year. The value of the combined $5 million stake of the three philanthropies has risen approximately 30-fold; even more so it has given the investors greater visibility into the industry.
Ford Foundation is also a limited partner in ex/ante, a venture capital firm focused on human agency and digital freedom, investing in technology that increases control over privacy, data, assets, and algorithms. Zoe Weinberg incubated ex/ante at Schmidt Futures, and has attracted a range of LPs, including Marc Andreessen and Chris Dixon of venture firm Andreessen Horowitz and fund of funds Cendana Capital.
“We know what the authoritarian surveillance state model of technology looks like. China and Russia have done a very good job perfecting that,” Weinberg tells ImpactAlpha. “We are interested in supporting the opposite.”
For Weinberg, the opportunity is as much about market structure as values: as power shifts from physical to digital infrastructure, the next generation of category-defining companies may be built around who controls data, networks and algorithms, and on whose behalf.
