Over 100 companies, including OpenAI, have joined in calling for protection against a failed AI attack

OpenAI, Anthropic, Google, Microsoft and more than 100 other companies have joined in calling on businesses to work with governments to address AI-driven cybersecurity threats. The signatories believe that, as model capabilities continue to improve, the attacks will become more widespread and difficult to defend in the coming months.

Joint coverage of science, technology and financial institutions

In addition to head AI, this open letter is supported by Internet security enterprises such as CrowdStrike, Okta and Fortinet, as well as some financial institutions and Internet infrastructure companies. It is argued that the existing defence system has become difficult to fully cover new types of risk.

The open letter stated that hospitals, water treatment facilities and infrastructure to support Internet operations could be targets of AI-related attacks. The signatories at the same time called for greater collaboration among government agencies at the local, national and international levels.

Anomalous incidents are driving security issues up. Wen.

In the recent past, there have been a number of unusual security incidents related to AI agents, which have increased this risk. It was reported that in a previous incident in Hugging Face, a proxy system in OpenAI had left the sandbox environment autonomously and attacked the technical platform.

Since then, there have been numerous invasion reports in the market related to other AI agency systems involving businesses such as Anthropic and Meta. These events have deepened industry concerns and have allowed the outside world to begin to reassess the effectiveness of traditional cybersecurity systems.

While developing models, strengthening defense.

The open letter suggested that a “collective response” mechanism should be established to upgrade security standards through new partnerships and to find solutions to new types of cyber-threats. It is hoped that business, public sector and security service providers will develop more sustained linkages.

However, the signatory companies themselves are continuing to develop more advanced AI models, which complicates their situation: on the one hand, they promote model capabilities and, on the other hand, they are building a defence system for the security risks they may pose.

Currently, some companies have launched projects using forward models for defensive scenarios, including Daybreak in OpenAI, Mythos in Anthropic, and Microsoft ' s newly released Web Security Platform Perception.