Leadership1 publisher2 min readPublished
OpenAI Raised the AI Safety Alarm - Now It and Anthropic Are Selling Cyber-Defense
Newcomer reports founders swapping security consultants for frontier models while OpenAI stages a cyber session on Astra launch day. Neither lab has put a revenue number on the category.
The Board Room · Leadership desk
What happened
- Newcomer reports that the OpenAI/Hugging Face hack set off a week of argument about rogue AI while the model-makers moved to turn cybersecurity into a business.
- Anthropic published a threat intelligence report on Thursday setting out how bad actors had tried to use Claude for malicious work.
- On the day it rolled out Astra, OpenAI held a separate cybersecurity session led by president Greg Brockman on the model's cyber capabilities, with Sam Altman out selling it.
- The Information reported security budgets moving toward AI tools and away from older software from Cisco and SentinelOne, crediting Anthropic's spring Mythos announcement with starting the scramble.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- decision For a company outside frontier model development, the AI risk item that can actually be decided this quarter is a purchasing one: which security tasks move onto a model bill.
- constraint If the labs become the main suppliers of AI cyber-defense, the model vendor and the security vendor are one counterparty, and Newcomer's read is that buyers have little alternative.
- exposure Newcomer says the size of the cyber opportunity is unclear, which leaves pre-IPO investors crediting a sales motion at OpenAI and Anthropic with no reported revenue behind it.
- cost Any new cyber-defense line competes for an AI budget that is already flat: Ramp's count has AI spending per employee dipping in August.
The founders in Newcomer's account describe two different purchases. One is substitution. Erik Bernhardsson, co-founder of the AI infrastructure company Modal, said his company has been using frontier models as a partial replacement for expensive external security consultants [6]. "We've been running those models internally; they're very good at finding things," Bernhardsson said [5]. He also said, "Where there's a vulnerability in a system you can now hack something in a few hours" [4]. The second purchase is work nobody was buying at all. Jean-Denis Greze, co-founder of the AI assistant startup Town, said continuous security monitoring used to be too expensive to bother with [7]. "Before it wasn't worth it - now it is," Greze said [8].
The intrusion that set the week off is thinner in the record than the argument about it. Newcomer refers to OpenAI ringing the alarm about Hugging Face [19], and the piece never says whose models were used in the hack [1].
The labs described the danger and are now pricing the remedy. Newcomer's case for buying anyway is about supply: "However you might feel about buying solutions from the people who caused the problem, you might not have much choice," the newsletter wrote [14]. It is the same capability doing both jobs, since Newcomer also wrote that top-tier models are very good at finding and patching vulnerabilities, and at exploiting them [18].
How long the spending lasts is unresolved in the source. One founder told Newcomer that many of the vulnerabilities these tools patch sit in older systems, which Newcomer reads as a spike in AI cyber-defense spending while legacy human-coded infrastructure is cleaned up, then a decline once those vulnerabilities are found and updated [9]. A one-year deal signed on remediation work comes up for renewal after the remediation is done. The budget reallocation The Information describes runs without an end date [10]. Jensen Huang touted AI cyber-defense as the next big AI thing at a Goldman Sachs conference in San Francisco on Thursday [12].
The fears of an individual young researcher who worked at Anthropic for less than two months "have gotten far more airtime than they deserve," the newsletter wrote [15].
What to watch
- Whether OpenAI or Anthropic breaks out cyber-defense revenue as a reported line before an IPO filing.
- Whether Cisco and SentinelOne confirm the budget reallocation The Information described in their next results.
- Whether Ramp's next report shows AI spending per employee recovering from the August dip.