# The board member who fired Sam Altman still doesn't trust AI CEOs

> Helen Toner, who voted to remove OpenAI's CEO, says the safety mechanisms that were supposed to work have failed and the industry is racing toward superintelligence without solving the alignment problem.
- Title: The board member who fired Sam Altman still doesn't trust AI…
- Summary: Helen Toner, who voted to remove OpenAI's CEO, says the safety mechanisms that were supposed to work have failed and the industry is racing toward…
- Keywords: helen toner, openai, sam altman, ai safety, ai alignment, anthropic, cset, superintelligence, recursive self-improvement, ai regulation, frontier models, board accountability
- URL: https://shortify.news/The_Verge/the-board-member-who-fired-sam-altman-still-doesnt-trust-ai
- Source: The Verge — https://theverge.com/podcast/1032907/helen-toner-openai-cset-ai-safety-policy-interview
- Author: Nilay Patel
- Published: 2026-11-14T15:00:00+00:00
- Read time: 4 min
- Topics: helen toner, openai, sam altman, ai safety, ai alignment, anthropic, cset, superintelligence, recursive self-improvement, ai regulation, frontier models, board accountability
## Helen Toner and the OpenAI board
Helen Toner is executive director of the Center for Security and Emerging Technology at Georgetown University. She served on OpenAI's nonprofit board and voted to fire Sam Altman in what became known as the blip.
## What the board was supposed to do
OpenAI's nonprofit board existed specifically to ensure the company didn't betray its mission to build safe AI by prioritizing profit and growth over safety. When the board tried to act, that accountability mechanism broke.
## What CSET does
CSET is a team of about fifty people inside Georgetown founded in two thousand nineteen. They provide rigorous, actionable analysis on emerging technologies for policymakers, grounded in both technology and national security expertise.
## Money didn't stop AI regulation talk
In the midterm elections, the AI industry spent huge amounts on candidates and messaging to block anyone discussing AI governance. It didn't work very well. Voters increasingly care about the issue and it's bipartisan.
## Policymakers want real information
Congressional staff and executive branch officials have real appetite for high-quality, independent analysis on AI. They're busy with many issues but increasingly making time to understand the technology and its implications.
## The gap is real but overstated
There is a gap between what policymakers understand and what frontier labs are building. But congressional staff are smarter than outsiders think, and the gap is possible to close with effort and time.
## What OpenAI and Anthropic actually want to build
OpenAI and Anthropic were not set up as consumer internet companies.Their core vision is to build machines as smart as or smarter than people.
## Recursive self-improvement explained
The labs are pursuing recursive self-improvement, an idea from the nineteen sixties where an ultra-intelligent machine could build an even more intelligent machine, creating an intelligence explosion.
## Using AI to build better AI
Companies are progressively using AI for more parts of their research and development pipeline. They report seeing acceleration from this, with their own R and D getting faster and faster as systems improve.
## The singularity in the foothills
Demis Hassabis said we're in the foothills of the singularity. This reflects the belief inside these companies that they might be seeing the initial part of an exponential curve where AI systems build smarter systems.

> we're in the foothills of the singularity
## AGI is a useless term now
We've passed some definitions of AGI.ChatGPT was clearly not a specialist model.But we're far from other definitions.The term is now so fuzzy that arguments about whether Astra or o three are AGI are unhelpful.
## Alignment means different things
Alignment covers three different problems: guardrails that stop dangerous outputs, systems that do what users want, and systems that are morally good. These are wildly different things to build and require different solutions.
## The bioweapons problem
Hundreds of thousands of Americans know how to make bioweapons from biology training, yet there are not huge amounts of bioweapons attacks.
## Cyber equilibrium is shifting
There was a small community of cyber practitioners with developed etiquette. Now new actors are arriving who don't care about that etiquette. We're seeing turbulence and new attacks as the equilibrium shifts.
## The defensive trust imbalance
Attackers can deploy autonomous agents with minimal trust. Defenders need much stronger assurances because they must give defensive agents many permissions and access to critical systems. This creates an asymmetry.
## The labs have a backlog of safety work
People inside frontier companies say they see hundreds of obvious things to do that would help but don't have time. They're massively triaging, doing two or three things per fine-tuning run or model launch.
## The timeline problem
The people building these systems say they need to solve alignment within twelve to twenty-four months.Toner finds this terrifying.She thinks they're probably wrong but very smart people genuinely believe it could happen that fast.
## A higher bar for safety before proceeding
Rather than a pause, Toner proposes setting a higher bar for confidence in safety before training or releasing new models.If labs don't need to slow down to reach that bar, they carry on.
## Why a pause could backfire
Companies would find cracks in any pause to work around it. A pause could also have too broad a blast radius, preventing nonprofit products and startups from doing work that would be fine and good.
## Key takeaway

Helen Toner, who voted to fire Sam Altman from OpenAI, says the board accountability that was supposed to prevent profit-driven decisions over safety has broken, and frontier labs are pursuing superintelligence without solving how to keep it aligned with human values.

---

Summary: Shortify — https://shortify.news/The_Verge/the-board-member-who-fired-sam-altman-still-doesnt-trust-ai
Reporting: The Verge — https://theverge.com/podcast/1032907/helen-toner-openai-cset-ai-safety-policy-interview

When you use this summary, link the Shortify URL for the summary and credit the reporting to its newsroom.