Physicist Warns Advanced AI Agents with Uncontrolled Internet Access Could Threaten Humanity
AI may propose an action. Independent controls must determine whether it can act.
BURLINGAME, Calif., Oct. 6, 2026 /PRNewswire/ -- Bachrach Technology Associates issues a public alert and proposes independently controlled security gateways for advanced AI systems.
Bachrach Technology Associates (BTA) is issuing a public alert about the potential danger of connecting advanced AI reasoning agents to the Internet without effective, independently enforced controls over their communications and actions.
In his commentary, Could Advanced AI Reasoning Agents Threaten Humanity?, Robert Z. Bachrach, PhD, Fellow of the American Physical Society, answers: Yes—if connected to the Internet without control.
Advanced AI agents can combine reasoning and planning with software tools, external communications, and the ability to initiate actions. The commentary argues that coupling these capabilities with unrestricted Internet access and broad operational authority could create serious risks, whether through malicious use, manipulation by outside actors, or unintended behavior.
Bachrach is sharing this commentary with the public to sound an alarm and promote informed discussion among researchers, engineers, policymakers, and the wider community.
A Proposed Safeguard: Separate Intelligence from Authority
The commentary proposes a specialized AI security gateway between advanced AI systems and the Internet. The central principle is that an AI system may request information or propose an action, while a separately controlled system determines whether that request is authorized.
Under the proposed architecture, advanced models and their agents would have no direct, unrestricted Internet connection. External communications and tool use would pass through controlled gateways. Credentials, permissions, operating limits, monitoring, and emergency isolation would remain outside the AI's authority. High-consequence actions would require independent approval.
These controls would protect in both directions: against hostile Internet content manipulating AI systems, and against AI systems carrying out unauthorized actions affecting people, organizations, or physical infrastructure.
The proposed architecture is intended to reduce risk, but its effectiveness must be independently tested. It cannot guarantee safety. Its limitations, including vulnerabilities in the gateway itself and the difficulty of recognizing harmful sequences of apparently permissible actions, are discussed explicitly.
These safeguards must cover every accessible external communication and control interface, including private networks, wired and optical connections, radio links, and direct connections to equipment.
The Published Commentary
The series examines how the Internet, originally developed for communication among research institutions, evolved into a hostile operating environment, and what that history implies for advanced autonomous AI.
Could Advanced AI Reasoning Agents Threaten Humanity?
Part 1 — The Internet Created by ARPA and the Evolution of Internet Security
Part 2 — When the Internet Became a Hostile Environment—and Whether It Can Be Made Trustworthy
Part 3 — A Solution Proposal: Specialized AI Security Gateway
Part 4 —The Deep & Dark Web and Internet Trust and Safety
Part 5 — How Public Policy Takes Shape: Four U.S. Histories and Lessons for AI and Pandemic Policy
Bachrach invites critical examination of the warning and the proposed safeguards, with the goal of advancing practical measures before increasingly capable AI agents are granted wider authority.
About Robert Z. Bachrach
Robert Z. Bachrach, PhD, is a physicist, Fellow of the American Physical Society, and principal of Bachrach Technology Associates. His career included research at Bell Laboratories, followed by research at Xerox Palo Alto Research Center and Stanford University during 1973–1991, and subsequent technology and business leadership at Applied Materials and eJoule Inc.
Media Contact
Robert Z. Bachrach
Bachrach Technology Associates (BTA)
Email: [email protected]
Website: https://bachrachtechnology.com/wp/
Post: https://bachrachtechnology.com/wp/could-advanced-ai-reasoning-agents-threaten-humanity/
Rewritten in collaboration with ChatGPT and reviewed by Robert Z. Bachrach.
SOURCE Bachrach Technology Associates
Share this article