AI Risk Score — Week of August 19–25, 2025
AI Risk Score: 64
Base Score: 50
- Risk-Increasing Signals
- Risk-Reducing Signals
Risk-Increasing Signals
Tech rally stalls as AI enthusiasm cools
Rotation out of mega-cap tech and doubts about AI’s near-term payoff signal fragility in AI-led valuations.→ +3 → 53
Source: Wall Street Journal
Concentration risk: “Tech stocks are sending a warning”
A handful of AI-heavy giants dominate the indices; signs of slowing breakthroughs (and a lukewarm GPT-5) amplify bubble fears.→ +3 → 56
Source: Financial Times
“Brace for a crash before the golden age of AI”
Historic “installation phases” often over-invest before productivity arrives; today’s AI build-out fits the pattern.→ +2 → 58
Source: Financial Times
Investors scrutinize software as AI threatens existing models
Fear of AI cannibalizing software revenue pushes investors to re-rate the sector.→ +2 → 60 Source: Bloomberg
Nvidia becomes a near-term market risk event
Earnings/roadmap updates are a proxy for AI demand; disappointment could jolt broader sentiment.→ +1 → 61
Source: Investor’s Business Daily
Reports of poor enterprise ROI (the “95% failing” drumbeat)
Press coverage of an MIT-linked analysis says most gen-AI pilots aren’t delivering measurable returns, fueling bubble anxiety.→ +2 → 63
Sources: Economic Times
Report
Deepfake CEO scams surge
AI-enabled executive impersonations drive multimillion-dollar fraud losses—an operational and compliance risk for every finance team.→ +2 → 65
Source: Wall Street Journal
Prompt-injection exploit (“PromptFix”) shows AI agents can be steered.
A fake CAPTCHA triggers hidden prompts that auto-fill addresses and credit cards in an AI browser—evidence of real-world agent abuse.→ +2 → 67
Source: The Hacker News
Rushing AI into SDLC = insecure code
New survey reporting four in five firms knowingly ship vulnerable code—and a third say ~60% of their code is AI-generated.→ +1 → 68
Source: TechRadar
Geopolitics enters the chat:
China guidance to curb Nvidia H20Chip access and AI supply chains face fresh political risk, pressuring cross-border AI roadmaps.→ +2 → 70
Sources: Financial Times, Financial Times follow-up
Risk-Reducing Signals
California bills to govern AI in employment decisions advance
Would impose guardrails on algorithmic hiring, pay, promotion, and firing—direct mitigation of automated harms (cost concerns remain).→ −2 → 68
Source: CalMatters
NIST moves to standardize AI security controls (SP 800-53 overlays)
New concept paper + community effort to codify AI-specific controls—concrete steps toward hardening AI systems.→ −2 → 66
Sources: NIST Announcement, NIST COSAIS Project
NIST CSF 2.0 quick-start draft: managing emerging cyber risks
Fresh guidance to operationalize risk management across orgs, including AI-driven threats.→ −1 → 65
Source: NIST Draft
UK ICO audits police facial recognition use
Active oversight and audits can curb high-risk deployments and set accountability precedents.→ −1 → 64
Source: UK ICO
Final AI Risk Score: 64
Why it landed here:
Markets are wobbling under AI-bubble jitters (valuation concentration, ROI doubts, and geopolitics). At the same time, concrete oversight steps—from NIST control overlays to ICO audits and California’s employment-AI bills—offer counter-pressure. Net-net, perceived and actual risk rose this week, but governance momentum trimmed what would otherwise have been a steeper climb.
Energy Note (100 W light-bulb equivalent)
Estimated energy to research and draft this post (LLM + browsing): ~0.02 kWh → about 12 minutes of a 100 W bulb.