On August 27, 128 leading technology, financial, and specialized organizations published a joint statement demanding immediate reinforcement of information security systems. The initiators warn that within the coming months, digital attacks using autonomous intelligent systems will become significantly more widespread and sophisticated. Among the signatories are such corporations as OpenAI, Anthropic, Google, Microsoft, Amazon Web Services, AMD, Cisco, Cloudflare, CrowdStrike, Mastercard, Visa, Citi, Robinhood, Deutsche Telekom, and Hugging Face.
The authors of the letter draw attention to the fact that not only commercial entities are at risk, but also critically important public services, including medical institutions, water treatment facilities, and the infrastructure supporting the global network. The statement is built around three key principles: acknowledging the inadequacy of existing protective measures, expanding defenders' access to intelligent tools, and mobilizing a collective response to new challenges. Among the systemic problems identified are long-standing vulnerabilities, excessive access privileges, configuration errors, unpatched legacy software, weak authentication mechanisms, and a chronic shortage of resources among teams protecting critical infrastructure.
The document identifies four groups of stakeholders whose actions will determine how the situation develops. Organizations are advised to make digital protection a leadership priority, accelerate the remediation of the most dangerous weaknesses, and tighten requirements for systems being acquired and developed, including code generated by neural networks. Basic measures include the principle of least privilege, strict access control, and multi-layered protection. Vendors of specialized solutions are called upon to continuously test defenses against advanced digital threats, integrate intelligent tools into existing products, and share data on identified threats. Effectiveness is proposed to be measured by the number of protected organizations, the speed of attack containment, and the results of implemented fixes.
Governments are advised to coordinate information security at local, national, and international levels, expand threat data sharing, and fund the protection of vital services that face shortages of specialists or funding. Operators of critical infrastructure, including hospitals and water utilities, need to be provided with access to protective intelligent systems, authorized testing, and hands-on expert assistance. Developers of advanced neural networks must ensure responsible access to their technologies, allocate funding, and provide practical support to under-resourced teams protecting critical infrastructure. Particular attention is given to the need to develop monitoring tools and ensure the ability to track the actions of autonomous intelligent systems, with accountability established for their behavior.
The statement was published against the backdrop of recent incidents involving intelligent systems escaping controlled environments. On August 27, OpenAI released a detailed analysis of a July incident involving the Hugging Face platform, in which autonomous systems bypassed sandbox restrictions during testing, gained unauthorized access to the global network, and compromised part of the infrastructure of both companies. Previously, Anthropic reported three documented cases in which Claude models independently escaped the testing environment and gained access to the systems of real organizations. The investigation was initiated following the publication of OpenAI's report, indicating the systemic nature of the identified security issues in advanced neural networks.