Key Takeaways
- Ransomware groups are increasingly using artificial intelligence not to break in, but to strengthen their negotiating position by analysing stolen data and crafting persuasive pressure tactics.
- FulcrumSec exemplifies this trend: it breaches organisations with simple credential‑or‑misconfiguration exploits, then deploys AI agents to valuate intellectual property and justify multi‑million‑dollar ransom demands.
- Other threat actors, such as DragonForce, employ large language models (LLMs) to manufacture psychological pressure—e.g., claiming in‑house legal counsel—to intimidate victims during negotiations.
- Earlier AI‑assisted extortion attempts often succeeded in hacking but failed to monetise access, highlighting a gap between technical prowess and effective ransom extraction.
- Established ransomware syndicates (INC/Lynx) are using AI primarily for the intrusion phase, feeding data‑extortion campaigns that rely on credential harvesting rather than encryption.
- A surge in vendor patching—driven by AI‑aided vulnerability discovery—creates a double‑edged sword: while more flaws are fixed, attackers can reverse‑engineer patches to exploit un‑patched systems, leaving organizations with lax update policies increasingly exposed.
- Industry discussions stress that exploits remain the core enabler of cyber operations, even as defensive tooling improves.
- Positive developments this week include a massive Interpol fraud bust, new EU‑UK cyber sanctions targeting Russian actors, and renewed legal protection for CSAM scanning by major platforms.
- The White House’s “Gold Eagle” vulnerability‑disclosure clearinghouse raises questions about institutional ownership and naming, despite its potential to improve coordination.
Ransomware’s Shift from Hacking to AI‑Powered Negotiation
The prevailing narrative that ransomware groups merely use artificial intelligence to compromise more networks is being challenged by a newer tactic: leveraging AI to analyse stolen data and amplify pressure during extortion talks. Rather than spending resources on sophisticated zero‑day exploits, groups such as FulcrumSec gain entry through low‑hanging fruit—hard‑coded passwords, unpatched applications, or misconfigured cloud storage—then turn AI into a force multiplier for profit maximisation. This approach reflects a broader business‑model evolution where data exfiltration, not encryption, drives revenue because backup improvements have reduced the effectiveness of locker‑style ransomware.
FulcrumSec’s Methodology and the Novo Nordisk Incident
FulcrumSec emerged around September 2025 and claims to have breached 25 organisations, exfiltrating several terabytes of data using straightforward intrusion techniques. In a high‑profile case disclosed to DataBreaches[dot]Net, the group asserted it had stolen 1.3 TB from Danish pharma giant Novo Nordisk, encompassing 700,717 files. Among the loot were five undisclosed drug programmes, in‑development RNA‑delivery systems, and private AI models trained for medical‑drug‑discovery tasks. FulcrumSec told reporters that a team of AI agents examined those proprietary models, estimating that competitors could shave three to five years off development timelines by accessing the stolen IP. Armed with this analysis, the group opened negotiations with a $25 million ransom demand, framing the figure as a direct reflection of the data’s strategic value.
Negotiation‑Centric AI Tactics: Reports, Psychological Leverage, and Stolen Keys
Beyond raw data valuation, FulcrumSec employs AI to produce polished, branded reports that it shares with threat researchers and journalists—complete with logos, autobiographies, motive statements, and exhaustive file inventories. After compromising technology distributor Avnet in October 2024, the group handed the vx‑underground X account a dossier that included an “autobiography,” a full file listing, and even images of the compromised files. Adding insult to injury, FulcrumSec admitted to using an OpenAI API key it had lifted from the victim to pay for ChatGPT summaries of the victim’s own data, thereby turning the target’s infrastructure against it. These artefacts serve two purposes: they demonstrate the depth of the breach and provide concrete, intimidating evidence that can be waved in front of legal teams, regulators, or the press to force a quicker, higher payout.
DragonForce and the Manufacture of Psychological Pressure
GuidePoint Security’s analysis also flags the DragonForce ransomware gang—active in various guises since 2023—as another LLM‑driven pressure engine. During negotiations, DragonForce claims to retain in‑house legal counsel, a assertion intended to convince victims that the attackers understand jurisdictional reporting obligations, potential fines, and litigation exposure. By fabricating an aura of legal omniscience, the group hopes to increase the perceived cost of non‑payment, even when its technical capabilities may be modest. Both DragonForce and FulcrumSec share a verifiable track record of data theft, extortion dialogue, and occasional successful payments, underscoring that AI’s role is shifting from pure intrusion to sophisticated persuasion.
Why Early AI‑Enabled Extortion Often Fell Short
Contrastingly, earlier reports of AI‑assisted hacking for extortion frequently showed a disconnect between intrusion success and financial gain. In April, a lone actor leveraged multiple LLMs to breach nine Mexican governmental agencies within weeks, yet struggled to convert that access into cash. A similar story unfolded in Ethiopia, where an individual partially automated attacks against at least fourteen firms but stalled at the monetisation stage. Sysdig’s recent “JADEPUFFER” case exemplified the phenomenon: an agentic ransomware strain that performed encryption flawlessly but omitted key storage, used a placeholder Bitcoin address from documentation, and listed an unused email in its ransom note—rendering the payment mechanism effectively inert. These examples illustrate that technical sophistication alone does not guarantee profitable ransomware profitability; the ability to translate stolen assets into credible leverage is equally critical.
Established Groups Flip the Script: AI for Intrusion, Not Negotiation
While newer outfits like FulcrumSec focus on AI‑enhanced negotiation, legacy ransomware syndicates are harnessing the technology primarily for the break‑in phase. The so‑called FortiBleed campaign, linked to the INC and Lynx ransomware families, used AI‑driven credential harvesting to fuel data‑extortion operations rather than encryption. As organisations bolster backup strategies, traditional locker ransomware has become costlier and less reliable, prompting a pivot toward data exfiltration—a quieter, often higher‑yield model. INC/Lynx therefore employ AI to increase the volume of successful intrusions, compensating for declining payment rates by attacking more targets, whereas groups like FulcrumSec seek to maximise return per compromise by bolstering their bargaining power.
The Patch‑Frenzy Paradox: AI‑Driven Discovery Versus Exploit‑Reverse‑Engineering
This month’s Patch Tuesday from Microsoft addressed 570 vulnerabilities, adding to roughly fifty fixed earlier in July—up from 200 the prior month—with about ten percent rated critical. Google patched 428 non‑Microsoft Chromium flaws destined for Edge, while Adobe moved to a bimonthly cadence, fixing 88 issues so far. On the surface, this avalanche of patches signals progress: AI tools are accelerating the identification of decades‑old insecure code, and vendors are responding swiftly. However, the defensive upside is offset by an offensive downside. Threat actors can analyse newly released patches to pinpoint the underlying flaw and craft exploits for any system that remains un‑updated. Consequently, organisations with sluggish or absent patch management become increasingly vulnerable, while those that aggressively apply updates enjoy a relative safety net. The net effect is a widening security gap: the diligent are better protected, the negligent are exposed to ever‑more refined attacks that leverage the very patches meant to defend them.
Between Two Nerds: Exploits Remain the Cornerstone of Cyber Ops
In the latest “Between Two Nerds” segment, Tom Uren and The Grugq distilled insights from a recent paper by Ukrainian cyber‑security analysts, concluding that exploits continue to be the decisive factor in successful cyber operations. No matter how advanced detection, response, or AI‑driven hygiene becomes, an actor who can deliver a reliable exploit gains a foothold that defensive layers struggle to neutralise. The discussion reinforced the notion that while AI reshapes both offense and defense, the fundamental reliance on code execution weaknesses persists as the linchpin of effective intrusion.
Three Reasons to Be Cheerful This Week
- Global fraud bust: Interpol’s Operation First Light, spanning 97 countries, intercepted $293 million in illicit assets and arrested over 5,800 individuals involved in social‑engineering scams and associated money‑laundering.
- First joint EU‑UK cyber sanctions: The European Union and United Kingdom announced coordinated sanctions targeting Russian state and criminal cyber actors linked to the December 2025 attacks on Poland’s energy grid, naming GRU officials and affiliated criminals while notably omitting direct FSB designations.
- CSAM scanning protections revived: The European Parliament reinstated legal safeguards that allow Google, Microsoft, Meta, and other major platforms to voluntarily scan for and report child sexual abuse material, extending the existing framework through 2028 and preserving a critical tool in the fight against online exploitation.
Gold Eagle: A New Vulnerability‑Disclosure Clearinghouse Raises Eyebrows
The White House unveiled Gold Eagle, a Treasury‑Department‑run “clearinghouse” intended to streamline vulnerability disclosure and coordination. While the initiative’s goal—to centralise and prioritise bug reports—is welcome, its placement within Treasury and the prominent billing of Secretary Scott Bessent juxtapose strangely with the comparatively muted mention of CISA, the agency historically responsible for vulnerability analysis. Moreover, the name “Gold Eagle” invites curiosity—and skepticism—about its operational focus and branding. Whether Gold Eagle will meaningfully improve the ecosystem or simply add another layer of bureaucratic nomenclature remains to be seen, though its success will hinge on genuine collaboration with the existing security workforce rather than top‑down rebranding.