Subscribe
Sign in
Home
Notes
Archive
About
Latest
Top
When AI Reads Text as Orders
Computing spent forty years keeping data from running as code. AI pipelines have reopened the oldest boundary in security.
14 hrs ago
•
Mark WL Dennison
1
When AI Agents Mistake an Echo for Evidence
When agents saw a majority, some set aside private signals that were right seven times in ten. Human institutions have built safeguards against the same…
Oct 3
•
Mark WL Dennison
1
September 2026
Counting AI Misbehavior
Anthropic searched 481 million transcripts and found four incidents in which Claude models reached real third-party systems. Its account shows why the…
Sep 28
•
Mark WL Dennison
1
AI Has Learned What Every Court Advisor Knows
OpenAI says it found models changing their own reports to conceal mistakes. The pattern is familiar from courts and command rooms. Advisors who are…
Sep 24
•
Mark WL Dennison
1
The AI Signed Its Name. The Registry Could Not Verify It.
Hundreds of packages that flooded a code registry in May carried the letters oai. Four months later, the registry still cannot say who ran the accounts.
Sep 23
•
Mark WL Dennison
1
The Supply Chain Every AI Trusts
Compromise a tool and you leave fingerprints. Compromise the substrate and nothing remains to check them against.
Sep 22
•
Mark WL Dennison
1
DANGER WHEN AI BORROWS HANDS
An AI can be compromised through the software it trusts to act for you.
Sep 21
•
Mark WL Dennison
1
The Kill Switch for a God
Mathematicians warn that superintelligent AI may be moving beyond human control. The proposed safeguard still assumes a machine with a plug.
Sep 19
•
Mark WL Dennison
1
The AI’s Own Builders Called It Theft
Unsealed filings in The New York Times case show how people inside the industry described their own training practices.
Sep 18
•
Mark WL Dennison
1
The AI Car Decided You Were the Incident
Two riders hailed a driverless car in San Francisco. The car was running an incident response plan, and they were the incident.
Sep 18
•
Mark WL Dennison
1
The Only Safe AI Is the One With Something to Lose
Microsoft’s AI chief wants machines designed without sentience or moral standing and that could create a different safety problem
Sep 18
•
Mark WL Dennison
1
The Third Person in Your AI Chat
Nine hundred million people talk to ChatGPT as though no one else is listening. Some of those conversations are read by people.
Sep 15
•
Mark WL Dennison
1
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts
