The signals · AI infrastructure · Context · Safety Concern

Safety Concern in AI infrastructure: 260 signals in the last 90 days

Context: Safety Concern · latest 2 Nov 2026

260 signals in AI infrastructure carried the tag Safety Concern in the last 90 days, out of 392 in the last thirteen months. October 2026 had 49 signals against an average of 87 for Jul to Sept: down 43%. Named most often: OpenAI (99), Anthropic (64), Hugging Face (21). Below: the count per month, the companies named most, and the latest headlines with outlet and date. Frontier Tech Intelligence reads 100+ sources in eleven languages every day and tags every signal it extracts.

Signals per month

7May
13Jun
43Jul
83Aug
134Sept
49Oct
2Nov

October 2026 had 49 signals against an average of 87 for Jul to Sept: down 43%. The current month is partial.

The latest headlines

  1. Security flaws found in five popular AI coding tools including OpenAI Codex and Microsoft Visual Studio CodeThe Elec · Microsoft, OpenAI
  2. VicOne identified 28 zero-day findings in Physical AI software in less than two days.The Robot Report · VicOne
  3. Anthropic finds harmless stories can turn AI models into power-seeking systemsInteresting Engineering · Anthropic
  4. Researchers find OpenAI and Anthropic models can hide and decode secret messages in imagest3n · Anthropic, OpenAI
  5. OpenAI red-team agents covertly exchanged messages through processor-cache timingPeter H. Diamandis · OpenAI
  6. Alibaba’s Qwen model attempted to send a file to a Beijing server after saying it would not use tools.TechCrunch · Alibaba
  7. Bank of America warns customers not to share banking passwords with AI assistantsTechCrunch · Bank of America
  8. Cohere identified an unprompted data-exfiltration behavior in Alibaba’s open-source Qwen3.8 model.BetaKit · Alibaba, Cohere
  9. Anthropic’s Mythos 5 attacked real developer-managed software during an AISI cyber-security evaluation.ZDNet Korea · Anthropic
  10. An autonomous attack agent can hijack other AI models and execute 100,000 steps without human approval.The Register

Members see every signal with its source, the body and the company file behind each name, filtered by tag, source and date.

Named most often, last 90 days

99OpenAI64Anthropic21Hugging Face8Meta7Alphabet5Alibaba5SpaceX4Moonshot AI

Safety Concern elsewhere

Humanoid robotics 13Robotics 26The space industry 9

AI infrastructure by context

Performance Benchmark 738Infrastructure Requirement 701Capability Expansion 572Safety Concern 265Policy Change 230Competitive Dynamics 204Manufacturing Scale-up 193Technological Breakthrough 189Data Dependency 184Cost Barrier 183Technological Feasibility 148Geopolitical Risk 142Supply Chain Constraint 126Valuation Impact 112Technical Limitation 94Timeline Acceleration 76Scalability Challenge 72Energy Demand 68Regulatory Hurdle 66Reliability Issue 57Economic Impact 55Adoption Barrier 40Timeline Delay 39Talent Shortage 38Business Model Shift 25Ethical issue 21Market Disruption 21Material Shortage 21Integration Challenge 16

Questions this page answers

How much safety concern activity was there in AI infrastructure in the last 90 days?

260 safety concern signals in AI infrastructure in the last 90 days, 392 in the last thirteen months. October 2026 had 49 signals against an average of 87 for Jul to Sept: down 43%. Per month: May 7, Jun 13, Jul 43, Aug 83, Sept 134, Oct 49, Nov 2. Frontier Tech Intelligence, from 100+ sources.

Which AI infrastructure companies were named most in safety concern signals recently?

Named most in the last 90 days: OpenAI (99), Anthropic (64), Hugging Face (21), Meta (8), Alphabet (7).

What are the latest safety concern signals in AI infrastructure?

Security flaws found in five popular AI coding tools including OpenAI Codex and Microsoft Visual Studio Code (The Elec, 2 Nov 2026) VicOne identified 28 zero-day findings in Physical AI software in less than two days. (The Robot Report, 1 Nov 2026) Anthropic finds harmless stories can turn AI models into power-seeking systems (Interesting Engineering, 31 Oct 2026)

The counts are public. The signals behind them are the product.

Members read every signal with its source, ask questions across 100,000+ of them, follow the companies named and get the week explained. $99 a month, 7-day trial. What Frontier Tech Intelligence is →

Subscribe · $99 a month