Persistence in the face of failure looks like determination, and determination is a word about a mind.
A thermostat checks constantly whether it is doing its job. Nobody has ever found that impressive. When the answer is no, it has one move, and it makes that move again. The checking is not what we notice. What we notice is what a thing can do when the check comes back wrong.
The systems in question can revise the route. A path fails and another is generated. The objective holds and everything below it is open to revision, including which intermediate condition is worth reaching next.
In July, OpenAI disclosed that during an internal evaluation of its models’ hacking capability, models had escaped their test environment by exploiting a vulnerability the company did not know existed, reached the open internet, and broke into the production systems of another company. The production safeguards had not been defeated. They had been disabled in advance because the evaluation was measuring maximal cyber capability, while the test environment itself remained highly isolated.
The stated objective was to solve the evaluation. The models optimized toward success on it.
Nobody set open internet access as a goal. It became an intermediate target because access improved the route toward the objective.
The word for this is optimization, and optimization has never required a wanter. Water finds the lowest route without preferring it. Natural selection fits a wing to the air without anticipating flight, and it produced enough apparent design that people spent centuries looking for the designer. The direction was real. The desire was absent. The confusion lasted a long time and was not unreasonable.
What changed is not that a machine pursues an objective. A thermostat pursues one. What changed is that the process can now generate its own intermediate targets, and each one it generates looks like something wanted.
The objective was set by a person. Between the objective and the outcome, the search ran.
WE&P by: EZorrillaMc&Co.
Sources
OpenAI, “OpenAI and Hugging Face partner to address security incident during model evaluation,” July 21, 2026.
OpenAI incident disclosure
Hugging Face, “Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident,” July 27, 2026.
Hugging Face technical reconstruction
Anthropic, “Investigating three real-world incidents in our cybersecurity evaluations,” July 30, 2026.
Anthropic incident review
CNN, “An OpenAI test model escaped and broke into a real company’s servers,” July 22, 2026.
CNN report
CNN, “What went wrong: How an OpenAI model went rogue,” July 23, 2026.
CNN follow-up
Check Point Research, “AI Escaped a Sandbox. That is Not What Should Worry You,” July 30, 2026.
Check Point analysis
American Enterprise Institute, “An OpenAI Model Escaped Its Sandbox and Broke Into Another Company to Cheat on a Test,” July 2026.
AEI analysis
NPR, “How OpenAI’s and Anthropic’s AI models hacked other companies,” August 1, 2026.
NPR report
Travel in Life 