Urgent.News

What's breaking now, across thousands of outlets.

AI

The Cyber Risk Discourse is Broken

Open-weights, ideology, and acknowledging trade-offs.

The Cyber Risk Discourse is Broken

The open discourse surrounding AI risks often presents a lose-lose situation, with policy decisions potentially harming American AI competitiveness while simultaneously increasing long-term cyber risks. The debate centers around open weight models, which are viewed as a significant threat by both the national security community and AI risk moderates.

On the other hand, Chinese companies continue to release open-weight models with strong cyber capabilities, making decisions based on their societal and government risk assessments.

A recent report by Anthropic examining the risks of GLM-5.3 as an offensive cyber tool has been criticized for failing to address crucial questions such as the consequences of banning open models due to cyber risks or the safety assessments of Chinese companies. These questions are essential for understanding the ecosystem perspective on current AI risks.

Policy decisions, which often result in trade-offs, must be made in the context of a complex puzzle, where the shape of risk can change but rarely leads to universally better outcomes.

The current discourse surrounding open and closed models for AI also reveals a disconnect between public information and classified briefings, with many people expressing concerns about the onslaught of open models, despite the fact that closed models have been responsible for most existing cyber attacks. It remains unclear whether open model weights or closed model APIs pose a greater risk, and a closer examination of the cybersecurity ecosystem may provide valuable insights.

Ultimately, the debate around open weight models and their role in cyber defense is complex and multifaceted, with the potential implications of banning open models far-reaching. It is essential to consider the impact of closed models on the offense-defense cyber gap and the need for meaningful time to develop mitigations that allow defenders to utilize strong AI models on their private infrastructure.

As the debate unfolds, it is crucial to recognize the complexity of the issue and the importance of making informed policy decisions.

Written by urgent.news from Interconnects's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at interconnects.ai →

More in AI

The Receipt: I Priced Every Token of One AI Agent Task

Everyone talks about AI costs in aggregate. Dashboards show monthly spend, per-model totals, team budgets. Nobody shows you a receipt. So here's one.

  • $0.051 raw inference cost for AI agent task
  • Fully-loaded cost of $0.23 includes hidden expenses
  • Context (knowledge base retrieval) consumes 43% of tokens

Spec, Hash or Guess: Can LLMs Keep Spain's Tamper-Proof Invoice Ledger?

This is a submission for the Kaggle Benchmarking Challenge From 2027, every invoice issued in Spain has to come from software that writes VeriFactu records.

  • Nine language models tested on VeriFactu record specification
  • Seven models accurately generated hash text for all records
  • Gemma 4 31B most cost-effective at $0.14 for 100% accuracy

More from Tuesday 6 October →