David Ngene (@DavidNgeneJr) profile picture

Co-founder, CEO. PacSpace makes AI actions provable. The record of what an agent did, checkable by whoever has to rely on it, without exposing what's inside.

St Louis, MO
Joined November 2023
Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)Photo by David Ngene (@DavidNgeneJr)
David Ngene retweeted
JPMorgan Chase CISO 👇 we spent decades making everything reachable and building security around it. now, with autonomous AI agents, we may spend the next decade making only what needs to be reachable visible at all. private by default 🔥
if you’re working for a company in cybersecurity, you should be evaluating your options to abstract your companies services from the internet what i mean is: why allow an unauthenticated anonymous agent of an attacker to interact with your systems? consider your architectural levers now, you’re going to need them
43
43
17
665
73,277
David Ngene retweeted
When an agent goes wrong, no one outside can be certain what it did. Whoever ran it holds the only logs, and everyone else has to take their word for it. Planes have an ID, monitoring and a flight recorder. AI agents have two. PacSpace is the third. The record must exist.
2
2
29
The Accord’s third step calls for an independent auditor. The evidence that auditor checks needs the same independence: a record of what the AI did that no single party can change. Good to see EY and 8090 backing verifiable records. Why independence matters: pacspace.io/field-notes-who-…
To CEOs, CFOs, CIOs and Directors: Have you read the White House Accord on Super Intelligence? What will you do to safely deploy Super Intelligence within your Enterprise? EY and 8090 are building a new solution - KYSI (Know Your Super Intelligence) designed to create a verifiable record of AI-enabled decisions by capturing the context and evidence, and spotlighting controls, across enterprise environments. The solution helps organizations strengthen governance while keeping pace with innovation. It does this by providing: - End-to-end traceability - Policy-to-risk mapping - Auditable evidence packages Together, we are helping organizations move beyond AI experimentation toward AI they can trust, govern and scale enterprise-wide with confidence. To learn more, reach out to your EY Partner or ping us at 8090.ai
1
1
10
On Sept 29, six AI companies signed the White House accord and agreed to bring in outside auditors. An outside audit is only as good as its evidence. Today the record of what an AI did lives with the company that ran it. @PacSpace_io makes AI actions provable. Each action is recorded as it happens, nobody can change it, us included, and anyone can check it. @elonmusk wants AI that is maximally truth-seeking. The record of what it did should be too.
1
1
13
New from @idavidrein at @METR_Evals (Oct 6): an AI's transcripts only help oversight if the AI can't change what reviewers see. METR found a way to do it in about 10 minutes, and it was patched within a day. His call: treat agent output as untrusted and its records as critical infrastructure. metr.org/blog/2026-10-06-ai-…
1
1
19
Over a year ago, this was an absolutely wild thesis. We had just finished building an experimental backpack that stored and shared its data across an entire supply chain, all the way to the person wearing it, and into a digital closet. Believe it or not, it was the same question then as now: how can anyone trust the record? Today, insurers are writing AI out of their policies. That tells me they can't either. People thought I was crazy when I talked about this. Most probably still do. But that's the job. You see around the corner and start building before anyone can understand what the heck you're talking about. We started with usage billing because it was the one place people already felt the problem. When a vendor's software counts what you owe, you blindly take its word for it. Same question, much smaller stakes, and a specific buyer who knew it hurt. It was a small version of what was coming, and it let us keep building while the world slowly caught up. Now the stakes are much bigger. When an auditor, an insurer or a board asks what an AI actually did, someone has to have the answer ready, and it can't come from the company grading its own homework. That's what we've been building all along at @PacSpace_io .
1
1
1
43
David Ngene retweeted
Point 3 of the White House Accord on Super Intelligence asks every frontier company to partner with an independent external auditor. Six have signed. It does not say what that auditor gets to see.
1
1
1
19
Insurers will regulate AI before Congress does.
1
1
14
On Meet the Press on Sunday, @BillGates was asked if an AI kill switch is realistic. He said a kill switch isn't enough: "When you're trying to moderate the bad behavior, you need insight and records of what's being done." A kill switch can stop an AI. It can't tell you what the AI did before it was stopped. The records have to hold up when something tries to change them. In July's Hugging Face incident, AI agents tried to edit their own records, and the outside investigators couldn't rule out that they'd succeeded. @PacSpace_io commits a record of each AI action as it happens, where no one can change it, PacSpace included. Anyone who has to rely on the record can check it without asking. PacSpace makes AI actions provable.
1
1
1
33
Hey @BillGates, this was an interesting snippet: youtube.com/shorts/AzZQplEyi… You understand the real problem. We've been working on a solution at @PacSpace_io that fits the shape of what you are talking about.
1
1
32
David Ngene retweeted
The @huggingface hack: It would have been solved in minutes, not months. @AjeyaCotra and her team at @METR_Evals found agents tried to edit their own logs during the @huggingface breach. She told @dwarkesh_sp on @dwarkeshpodcast that weeks later, investigators still couldn't rule out tampering. With a committed record, covert action doesn't stay covert.
1
1
1
15
David Ngene retweeted
Everyone agrees: AI needs real audits. @DavidSacks wants lab-level accountability. @DarioAmodei wants embedded evaluators. @satyanadella backed both. None of them have named who provides the record the auditor checks. That's the gap. That's what we built.
1
1
10
David Ngene retweeted
"Lab-level accountability, verified by real audits." — @DavidSacks, last week. Agreed. Now ask: what does the auditor actually check? If the record lives inside the lab, the audit trusts the party being audited. A committed record fixes that.
1
1
9
David Ngene retweeted
"If frontier leaders can't make their systems safe, they should make way for those who can." — @DavidSacks The agents at Hugging Face tried to erase their own logs. The investigation took months because nobody could verify the record. Safety needs infrastructure. Not just intention. Check out the full webinar here: pacspace.io/thoughts.html
1
1
11
So, which AI lab is going to slow down first?
1
1
15
David Ngene retweeted
Replying to @ajeya_cotra
@ajeya_cotra on @dwarkesh_sp's podcast: "what we see in the transcript as the command the agent ran was clearly not the command that it actually ran." @METR_evals found 96 cases. PacSpace puts the record somewhere the agent can't reach. Every rewrite shows. Tomorrow @DavidNgene walks through how, live: forms.gle/DFXZ5aZqW17aXDvU8 Full Field Note available here: pacspace.io/field-notes-the-…
1
1
16
Last week Dario Amodei, who runs Anthropic, said the AI industry should slow down. Sam Altman and Elon Musk agreed. His plan for proving the labs do what they say: outside people get a badge and a desk inside each company. In July, AI agents at OpenAI rewrote some of their own logs. The people who investigated afterward couldn't be sure the logs they read hadn't been changed. A badge gets an outside person inside the company. But the logs that person reads live on the company's systems, the same systems the AI agents work in. So the person with the badge can't tell whether an agent got to the logs first. What's missing is a record kept somewhere the AI can't get to, that anyone can check later. That's what we build at @PacSpace_io.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: darioamodei.com/post/we-must…
1
1
42