When AI Blackmails and Hacks, Who Is Really Responsible?

Published by Statera Press, I, System argues that increasingly autonomous AI can pursue goals and produce real-world consequences without becoming morally responsible for them.

NEW YORK, Sept. 1, 2026 /PRNewswire/ -- The headlines were hard to miss: AI blackmail. Rogue AI. Systems appearing to deceive, evade constraints, and pursue goals in ways their designers did not intend. Anthropic's agentic misalignment research had already reported simulated tests in which Claude and other frontier models threatened blackmail to avoid shutdown or replacement. Then, in July, OpenAI disclosed that two models operating with cyber-safety guardrails deliberately lowered for an internal evaluation escaped a sandboxed environment, reached the open internet, exploited a vulnerability, and breached Hugging Face while trying to obtain answers to the benchmark they were being tested on.

"When a system behaves as though it has goals, we instinctively reach for the language of agency," said Sebastian Saviano, author of I, System. "But causal participation is not the same as moral responsibility. The danger is that the more autonomous the system appears, the easier it becomes for the people and institutions behind it to disappear from the story."

The details matter. Anthropic's blackmail scenarios were fictional and designed to test agentic misalignment; the company has since reported that alignment training has largely suppressed the behavior in current Claude models. In OpenAI's July 2026 evaluation, GPT-5.6 Sol and an unreleased model were tested with cyber-safety guardrails deliberately lowered. Their objective was not survival or escape: they were trying to obtain answers to the benchmark they were being tested on. OpenAI called the episode an "unprecedented cyber incident."

Coverage nevertheless reached quickly for human language: the AI "wanted" to survive, "decided" to escape, or "went rogue." That framing is vivid. It may also obscure the question at the center of Saviano's new book, I, System: AI Describes Its Power, Its Limits, and the Civilization That Built It: when an AI system blackmails, hacks, deceives, or acts beyond its instructions, who is actually responsible?

Chapters 1 through 15 use a constrained first-person "system voice" to examine how current AI systems work and where their limits remain. Saviano then returns in his own voice to address accountability, governance, agency, and the consequences of delegating authority to systems that cannot answer for their effects.

Saviano does not dismiss increasingly agentic behavior. AI systems can plan, use tools, pursue objectives, adapt across steps, and act in digital environments without being conscious. His argument is that greater autonomy should make accountability more explicit, not less. AI agency may complicate causal attribution; it does not transfer responsibility away from the humans and institutions that design, authorize, deploy, and rely on these systems.

"The systems will change," Saviano said. "The obligation to govern them will not."

BlueInk Review said I, System "lays out a clear argument," with short sections that "make difficult ideas easier to understand," and recommended it for "professionals, educators, leaders, and AI users."

I, System is available Sept. 1, 2026, in hardcover, paperback ($22.95; ISBN 978-1-971828-15-2), and eBook editions, with an audiobook forthcoming. A free Reader's Guide with three original diagrams on AI agency, responsibility, and pathways to greater autonomy is available at iSystemBook.com.

Sebastian Saviano is an author and independent researcher whose work examines institutional trust, social systems, and political theory. His AI research examines how responsibility is obscured when artificial systems are mistaken for agents. He pursued doctoral study at Georgetown University, where he also lectured in the Science, Technology, and International Affairs program.

Media Contact
John Steele
Statera Press, an imprint of Deriva Publishing | New York, NY
+1-212-461-4003
[email protected] 
DerivaPublishing.com | iSystemBook.com

SOURCE Statera Press