OpenAI says its new security model, GPT-6 Cyber, helped its researchers build working exploits for 948 flaws in the Linux kernel and find 81 previously unknown high-severity bugs in Firefox. The figure that needs the most care is 948, because it is the last number in a longer process, not a count of new discoveries.
OpenAI's own posts couldn't be checked directly for this story, so every figure below is attributed to the outlets that reported them. Those figures largely trace back to OpenAI, and outside confirmation so far is limited.
How the Linux work ran
TechRadar describes a five-stage pipeline: discovery, dynamic validation, exploitability assessment, patch generation and human review. Each stage narrows what the last one produced.
First, AI agents picked out security-relevant parts of the kernel and scanned them, and TechRadar reports that more than 2,300 candidates emerged within 12 hours. BleepingComputer puts the exact count at 2,308.
Next, further agents wrote proof-of-concept code to test each candidate. TechRadar says the system reproduced 1,724 findings, in a process that lasted 30 hours.
The agents then ranked those findings by impact. TechRadar says this produced 1,169 working local privilege escalation exploits, which fell to 948 distinct flaws once duplicates and false positives were removed. OpenAI then built patches for them. TechRadar puts the cost of the whole effort at roughly $1.21 million in API credits.
Local privilege escalation has a specific meaning here. BleepingComputer says these exploits could let an unprivileged user gain root access. An attacker would need to be on the machine already, so this is a different risk from a remote takeover.
What 948 does not mean
The 948 are not all new. The Register reports that 588 of them were previously unknown. TechRadar adds that maintainers had already patched 61 by the time OpenAI finished its analysis, that another 43 had staging patches, and that just under 300 were already being reported and discussed publicly.
TechRadar says OpenAI's researchers reviewed the exploits and patches before contacting kernel maintainers, and that the remaining findings are still being addressed. The Register says the findings are going to maintainers with potential patches for review. None of these reports quotes a maintainer.
One more detail is worth noting. TechRadar says OpenAI's Linux report never names the model, calling it only an "upcoming cyber model". The Register and BleepingComputer both attribute the 948 exploits to GPT-6 Cyber in their reports on the model's release.
Firefox
The Firefox figure is also OpenAI's. BleepingComputer reports that Mozilla has already released fixes for many of the 81 bugs, and The Register says Mozilla has started fixing them. Neither report quotes Mozilla on the count.
Who can use it
General users can't. The Register says access starts with approved customers in Daybreak Red, OpenAI's program for offensive security work, through the API and Codex, and will widen to all Daybreak Red customers over the coming weeks. BleepingComputer adds that a separate program, Daybreak Blue, lets verified defenders investigate vulnerabilities and validate fixes with fewer restrictions.
OpenAI rates the model "Critical" under its Preparedness Framework, which The Register describes as its highest-risk tier for cyber capabilities. BleepingComputer says that is up from "High" for the previous model.
The Register also reports the benchmark OpenAI is using to show progress: 48.5 percent on ExploitGym, against 32.1 percent for GPT-5.6 Cyber. OpenAI says the model avoided exploiting out-of-scope services over 99 percent of the time, with 0.07 percent of cases going beyond the intended scope. It declined to tell The Register what happened in those cases.
BleepingComputer quotes OpenAI's reasoning for releasing it at all: "It is clear that we have a limited window to use frontier models to secure software."
Reporting notes
