Field report SE-86 / L7 · sheet /macro
Boardroom MacroFiled from the territory
Bridgewater Prices a One-in-Three Chance of Economy-Destabilizing Cyberattack
Bridgewater is positioning for a one-in-three chance of an economy-destabilizing cyberattack within a year, as zero-day exploits roughly double since open-source hacking models emerged.
- Grid
- SE-86 / L7
- Surveyed
- November 20, 2026
- Reading
- 3 min read
- Correspondent
- Olivia Hart

Field notes
- Bridgewater assigns a one-in-three probability to an economically destabilizing cyberattack within the next year.
- Zero-day exploits have roughly doubled since open-source models with real hacking ability first appeared.
- Exploitation lag after disclosure has fallen to 11 days from about 30.
- GLM 5.3 costs $1.40 per million tokens versus up to $10 for OpenAI and Anthropic models.
- The FBI warned on Nov. 13 that hackers using the ARTEX framework reached privileged access in under three hours.
Bridgewater Associates is managing money on a one-in-three probability that an economically destabilizing cyberattack lands within the next year. The Westport, Connecticut hedge fund laid out the position in its Nov. 13 Daily Observations note, tying the risk directly to the spread of AI-enabled offense.
The note drew a pointed coincidence: the release of GLM 5.3, an open-source model from China's Z.ai, came alongside a surge in cyberattacks, according to MarketWatch. The analysis did not examine why attackers favor GLM. Pricing data on OpenRouter offers one clue — GLM costs $1.40 per million tokens, while comparable models from OpenAI and Anthropic can run up to $10.
What did Bridgewater actually build?
To track the threat, the firm constructed a time series of cyber incidents and paired it with a framework that links each episode to its underlying causes, plus a live pipeline that follows new exploits. The finding: incidents are being logged far more often, and a growing share is attributed to AI, with exponential growth continuing.
Bridgewater described AI-enabled cyber warfare as spreading at an exponential pace with few checks — spreading, as its analysts put it, much as a virus would.
What does the attacker-defender balance look like?
Two of the firm's analysts, Hudson Attar and Arham Habib, framed the situation as a race between machines. "We are in the middle of an arms race where malicious actors are using agents to find exploitable bugs and defenders are using agents to preemptively find and patch vulnerabilities," they wrote. "While models have significantly accelerated work on both sides of the ledger, the defenders have the much heavier burden of patching all bugs before they can be exploited."
The numbers back that asymmetry. Zero-day exploits — attacks that give defenders no advance warning — have approximately doubled in number since open-source models with real hacking capability first appeared, Bridgewater said. The typical lag from vulnerability disclosure to exploitation has shrunk to 11 days, down from about 30 previously.
Who else is exposed?
The model itself has crossed into mainstream infrastructure. Amazon, Google and Microsoft host GLM models, and Hugging Face used GLM 5.2 to defend against a breach attempt by an OpenAI rogue agent — the same family of tools now cutting both ways.
Open-source agent frameworks have drawn separate warnings. On Nov. 13, the FBI cautioned that hackers relied on a framework called ARTEX to strike organizations worldwide, in some cases needing under three hours to move from first targeting to privileged access, Axios reported.
"We're seeing more active attacks than ever right now," Sam Rubin, senior vice president of Unit 42, the threat intelligence and consulting arm of Palo Alto Networks, told Axios.
How are markets responding?
The trade has already started. The Amplify Cybersecurity ETF jumped 6.1% on Wednesday, the same day Palo Alto Networks raised its annual outlook.
Policy pressure is building in parallel. In August, more than 100 technology companies — including OpenAI, Anthropic, Google and Microsoft — signed an open letter urging a unified defense against AI-enabled cyberattacks. The letter warned such attacks would grow more widespread and sophisticated as models become more capable.
For practitioners, the practical takeaway is speed: an 11-day window from disclosure to exploitation leaves little room for manual patch cycles, and cheap offensive models have removed the budget barrier that once limited attackers.
via qz.com (Source)
More from Olivia Hart
Show full bio
News editor covering regulation, policy and their impact on corporate strategy and chief-executive leadership at CEO Frontier.
510 articles
Adjacent sheets
- Three Top AI Security Models Agreed on Just 1 of 121 Vulnerabilities
- AI-Patched Software Is Likely Behind This Week's Outage Wave
- Ex-OpenAI Researcher: AI Accident Is the Most Likely Cause of Your Death
- OpenAI Gives NATO Its Daybreak Cybersecurity AI Models
- Cloudflare Cuts 20% of Workforce in 'AI-First' Restructuring