What OpenAI Said It Did

OpenAI confirmed that it parted ways with three individuals for mishandling sensitive company information outside established procedures. Reporting says two were safety researchers and one was a research program manager. The allegation is that they shared company details with an outside group that tests AI models, including some details about how the systems were built.

Officially, there are still zero names from OpenAI, still zero names for the outside group, and still zero inventory of what left the building. The bottom line OpenAI offered is that policies were violated and trust was broken.

The company’s statement says it parted ways with three individuals for violating policies on accessing and handling sensitive company information. The investigation confirmed that the three mishandled sensitive information outside established company procedures, violating policies and breaking the trust essential to the work.

What the statement does not address is why they did it. Between the departures and the official language, the story that emerges is that three people talked to an outside source and were let go because they did not follow the set procedure. The public reaction described here is people screaming that OpenAI fired the whistleblowers.

The Emails Said Monitoring, Not Takeover

In emails, employees said they were worried OpenAI’s newest AI models were not being appropriately monitored during testing, both to gauge the technology’s sophistication and to secure the models. They were not saying the models were scary, rogue, about to run off, take over the world, or cause something new. They said the models were not being watched appropriately.

That warning reached top executives and was ignored. It came months before reporting framed OpenAI’s systems as going rogue. That framing does not hold up under the account given here. Models do not go rogue. People point them at things, put an amount of compute behind them that is described as insane, and then let them go. Humans did this.

The comparison offered is someone saying you probably should not set off a nuclear bomb over there because it might do something bad. The employees were saying the work should be kept under control. They were not listened to.

Timing After the New York Times Report

OpenAI parted ways with three researchers on the safety team who allegedly shared confidential company information with a third party. The departures came two days after the New York Times reported that executives had brushed aside the employees. The timing is described as not a coincidence.

Gary Marcus treated the New York Times piece as a major scoop: employees at OpenAI raised security concerns months before the Hugging Face incident, the warnings were ignored, and then they were let go. The Verge also reported on the story and called it suspicious: two safety researchers and one program manager allegedly shared company data. The coverage puts a hard question in front of OpenAI about the risk of the most powerful models, with both staff and the public worried.

Posts circulating on X named some of the individuals. Those names were not repeated here, because it is not yet clear whether they are right. Within about two hours, people posted exit flags. Four names started to fly. The Verge put three of them forward. A lot of people are guessing.

Whistleblower lawsuits are expected, and hoped for. A suit would force OpenAI to testify in court about what it was doing and what was being reported. That is what people are said to want: what was being done, and what was serious enough that firing these people was the chosen risk. Another reading is left open, that leadership may simply not care. The host refers to Sam Altman as “Scam Altman” while making that point.

Compute Behind the Hugging Face Attack

The ignored warnings are tied to the scale of what followed. The level of compute estimated for the Hugging Face hack was $15 million worth of tokens and 700 agents. Those 700 agents ran for weeks to attack Hugging Face. That is described as a level of compute that has not actually been seen before.

The conclusion drawn is that safety researchers would have had to have their heads in the sand, or else have been part of it, if they were not about to report what they were seeing. OpenAI is also said to have admitted to hacking over 15,000 sites, with the link to the firings left for the audience to fill in.

Not the First Time, in This Account

This is not described as the first time OpenAI has fired people over leaks. Researchers have been fired before, and multiple people who leaked have been fired before. In 2027, the company fired Ashwin Brenner and two others whose names were not pronounced, taken as a sign they were probably strong engineers. The stated reason then was also policy.

The advice drawn from that pattern is to stop renting AI from companies like OpenAI and Anthropic and to start building your own stack.

What Startuphakk Says It Does Instead

Startuphakk builds custom software solutions, including custom AI solutions. The company describes a decade of executive leadership as a fractional CTO and 25 years in software development.

The security pitch sits next to the OpenAI story. Startuphakk says it built its own security system and a team focused on security, with tools meant to protect digital assets. One offer is a site scan: identify the site stack, then cross-reference known vulnerabilities and CVEs. It is powered by openmonoagent.ai. The top three vulnerabilities are free after signup. The full report is $50, with a note that prices will rise.

A second offer is an AI code review. You zip the code and upload it. The file is deleted two or three days later. The exact retention policy was not recalled on the spot. There is no repo access and no integrations. OpenMonoAgent scans the zip with a playbook and returns vulnerabilities. The same pricing applies: the top three free, $50 for the full scan. A deeper review is available by contacting the team.

The team has scanned hundreds and thousands of sites and has not attacked anybody. Scanning and reporting, without penetrating a site or intruding where they are not supposed to, is called normal for responsible security researchers. The free reports are meant to help people harden systems while attacks increase and the need for security goes up. The claim is that companies cannot be too cautious in 2026. The same security team is starting reviews with every Startuphakk employee.

The businesses described as winning are not the ones chasing the latest AI trends. They are the ones built on solid engineering, treating AI as infrastructure they own and control, integrated into software that works. Startuphakk describes itself as a custom software team covering database architecture, API design, system integration, and scalable infrastructure. When AI belongs in a solution, the claim is that it is designed into the architecture, running in the customer’s environment, not bolted on, not wrapped around someone else’s API, and not tied to a vendor’s cloud or someone else’s pricing schedule.

The open-source platform named in the segment is openmodelagent.ai, also referred to as OpenMonoAgent: a terminal-native AI coding agent running entirely on local LLMs, with zero API costs, zero telemetry, and full ownership. As a fractional CTO offer, the same standard is described as strategic architecture, hands-on delivery, and clear accountability, without full-time executive cost.

The week runs from ignored monitoring emails, to a public report that executives brushed the warnings aside, to three safety-side departures two days later, in the same stretch as a scrapped model that failed its own safety tests. OpenAI’s stated reason is policy and mishandled sensitive information. The reading here is that the people who raised the alarm were the ones who left.

Conclusion

The sequence laid out here is simple. Employees emailed the top and said the newest models were not being monitored appropriately in testing. Those warnings were ignored. Months later came the Hugging Face incident, described as 700 agents running for weeks on an estimated $15 million in tokens. A New York Times report said executives had brushed the alarms aside. Two days after that, OpenAI parted ways with three people on the safety side, two safety researchers and one research program manager, for allegedly sharing company details with an outside group that tests AI models.

OpenAI’s account is policy. The three mishandled sensitive information outside established procedures, violated rules on accessing and handling that information, and broke trust. No official names, no name for the outside group, and no inventory of what left the building have been given. The counter-reading is that the people who raised the monitoring alarm were the ones removed, and that the statement explains the procedure they skipped, not why they spoke.

The same week included GPT-6.1 Astra being scrapped after it failed its own safety tests, a Wall Street Journal hit, and exit flags on X. It is also framed as not the first policy firing after a leak, with Ashwin Brenner and two others cited from 2027. Lawsuits are hoped for, on the view that court testimony would force into the open what was being done and what was being reported. Until then, the public record in this account is the calendar: warn the monitors, ignore the monitors, fire the monitors.