Retired Lt. Gen. Mary O'Brien calls AI-fabricated outside data a bigger threat than exfiltration
Retired Lt. Gen. Mary O'Brien told CISOs that AI-made external information reaching executives is now a larger threat than exfiltration. Her examples include the Cracker Barrel backlash, where she did not claim fake accounts caused the stock decline.
The Watch · Security desk
Key figures: 21% of the roughly 3,200 profiles driving the Cracker Barrel conversation on X were fake, per Cyabra data cited by O'Brien. Content from the fake accounts reached 4.45 million potential views.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Profiles driving the X talk that were fake | 21% | 10 | |
| Profiles driving the X conversation, roughly | 3,200 profiles | Cyabra data cited by O'Brien | 10 |
| Fake-account content reach | 4,450,000 potential views | 11 |
What happened
- Cyabra research she cited counted 37,000 AI-generated or manipulated items in the first week of the recent Iran escalation, drawing 145 million views.
- In August, a mayoral campaign reposted invented polls from an unidentified firm, and within minutes they had shifted prediction-market odds, with no one having verified them first.
- Cyabra data on the August 2025 Cracker Barrel logo backlash put fake accounts at 21% of roughly 3,200 profiles driving the conversation on X.
- O'Brien reduced the problem to three questions for each external source: who owns the source, who has verified it, and whether anyone in the room would catch it being wrong.
Why it matters
- exposure In O'Brien's account, executives who act on outside polling, sentiment or social data have no check in the security program that tests whether that input is real.
- cost The work lands on teams Eftekhari describes as already stretched by IAM, patching and training, and he says it sits outside what a security organization is responsible for.
- decision A board can adopt the ownership questions now. They need a named owner and verifier per source, and the account describes no product to buy.
O'Brien's case starts from a definition. "We've built zero trust for identities, devices and code. We still extend implicit trust to information," she said [3]. By her account, a fake poll sitting on a fully secured server clears every integrity check a security program can run [5]. "Integrity means not altered. It has never meant true," she said [4].
She described a pattern with three intents: gray-zone signaling, division and economic effect [8]. The Iran-related content pushed specific narratives at Pentagon leaders, industry decision-makers and the public [7]. "It wasn't random," she said [7].
The per-item figures are small. The 145 million views across 37,000 items average about 3,900 views per item [20]. In the Cracker Barrel case, the 21% fake share works out to about 670 accounts [19]. Those accounts put out over 1,000 pieces of content, with a potential reach of 4.45 million views [11]. The board ultimately reversed the logo decision [12].
Parham Eftekhari joined O'Brien on stage. "Organizations are being targeted in ways that fall outside of the traditional scope of what a security organization is responsible for," he said [14]. He is an executive vice president at CyberRisk Alliance, the parent company of SC Media, which published the account [17]. He said poisoned external data and misinformation used to cause harm or fraud rarely come up in his work with CISOs [15]. His proposed rehearsal is to war-game a brand decision under a manufactured backlash before it happens, not during [16].
In our view the record supports a narrower claim than the headline one. Outside inputs to executive decisions have no assigned owner, and O'Brien's ownership questions give a way to assign one [13]. Whether the inbound threat is larger than exfiltration is untested here. The account includes no loss figures for either. The Cyabra counts she cited come from a firm she advises [6].
What to watch
- Whether Cyabra or O'Brien publish how the 37,000 items were classified as AI-generated or manipulated.
- Whether the firm behind the fabricated August polls, or the mayoral campaign that reposted them, is identified.
- Any loss figure that compares inbound fabrication with exfiltration, the comparison the ranking claim needs.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap+15
- Incentives55
- Confidence60
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Speaking last week at the CyberRisk Collaborative NYC Leadership Exchange, retired Lt. Gen. Mary O'Brien, who spent 34 years in Air Force intelligence and cyber operations, said her mission of keeping data in has reversed.
- [2]
O'Brien said exfiltration is still a risk but the larger threat is bad external data, created at scale with AI, reaching executives who act on it instead of on their own trusted internal analysis.
- [3]
O'Brien said: "We've built zero trust for identities, devices and code. We still extend implicit trust to information."
- [4]
O'Brien said: "Integrity means not altered. It has never meant true."
- [5]
O'Brien noted that a fabricated poll stored on a perfectly secured server passes every integrity check a security program runs.
- [6]
O'Brien cited research from Cyabra, a firm she advises, identifying 37,000 pieces of AI-generated or manipulated content in the first week of the recent escalation with Iran, drawing 145 million views.
- [7]
O'Brien said the Iran-related content pushed specific narratives aimed at Pentagon leaders, industry decision-makers and the public, and said: "It wasn't random."
- [8]
O'Brien described three intents behind such campaigns: gray-zone signaling, division, and economic effect.
- [9]
In August, fabricated polls from an unknown firm were reposted by a mayoral campaign and moved prediction-market odds within minutes, before anyone checked whether the polls were real.
- [10]
Using Cyabra data on the August 2025 backlash to Cracker Barrel's logo change, O'Brien said that of roughly 3,200 profiles driving the conversation on X, 21% were fake.
- [11]
The fake Cracker Barrel accounts produced more than 1,000 pieces of content that reached 4.45 million potential views.
- [12]
Cracker Barrel's stock decline coincided with the amplification, O'Brien noted, without claiming it caused it. The board ultimately reversed the decision.
- [13]
O'Brien said the case comes down to three questions for every external source a consequential decision depends on: who owns it, who verified it, and who in the room would know if it were wrong.
- [14]
Parham Eftekhari said: "Organizations are being targeted in ways that fall outside of the traditional scope of what a security organization is responsible for."
- [15]
Eftekhari said the use of poisoned external data and misinformation to cause harm or fraud is something that rarely comes up in his work with CISOs.
- [16]
Eftekhari argued for war-gaming a brand decision under a manufactured backlash before it happens, not during.
- [17]
Eftekhari is EVP of Communities at CyberRisk Alliance, the parent company of SC Media, and Chairman of the Institute for Critical Infrastructure Technology.
- [18]
According to Eftekhari, security teams already struggling with basic hygiene such as IAM, patching, security awareness training and data protection are now consumed by reducing AI-related risk.
- [19]
About 670 of the roughly 3,200 profiles in the Cracker Barrel conversation were fake.
- [20]
The Iran-related content averaged about 3,900 views per item.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- Cybersecurity GovernanceFollow
- Zero trustFollow
- AI-generated disinformationFollow