The most likely way for the world to be destroyed, experts agree, is by accident. That's where we come in; we're computer professionals. We cause accidents. (Nathaniel Borenstein)
SAFETY
-
AI Apps Creating Non-Consensual Intimate Images of Women
By
–
AI Apps Are Undressing Women Without Consent And It’s A Problem #AI #nudification #apps are making it frighteningly easy to create #fake sexualized images of women and teens, sparking a surge in #abuse, #blackmail and #onlineexploitation. This article explores the #dangers of
-
Comprehensive Taxonomy of LLM Hallucinations: Intrinsic vs Extrinsic Errors
By
–
8. A Comprehensive Taxonomy of Hallucinations Presents a detailed taxonomy of LLM hallucinations, distinguishing intrinsic vs extrinsic errors and factuality vs faithfulness, and covering manifestations from factual mistakes to domain-specific failures.
-
GPT-5 Admits Uncertainty Instead of Hallucinating
By
–
GPT-5 admits it "doesn't know" an answer! This is one of the huge improvements over previous models: instead of hallucinating, it lets you know its limits.
-
Prompt Injection Security: Design Patterns and Prevention
By
–
Not that let you use all three legs of the lethal trifecta at the same time For prompt injection generally I like these two papers: https://
simonwillison.net/2025/Apr/11/ca
mel/
… and https://
simonwillison.net/2025/Jun/13/pr
ompt-injection-design-patterns/
… -
Agent Reliability: Catching Hallucinations and Verifying Groundedness
By
–
Agent Reliability A practical guide to help you catch hallucainations, verify groundedness, and monitor tool usage for LangChain/LangGraph applications: https://
docs.futureagi.com/cookbook/cookb
ook13/Adding-Reliability-to-Your-LangChain-LangGraph-Application-with-FutureAGI
… -
LLMs Cannot Distinguish Context: Collaboration vs High-Stakes Evaluation
By
–
Two situations: 1 I wave a co-worker to my monitor to show them a file I have open. "Is this right"? 2 I sit down someone at a table. They have 2 hours to respond. This is an exam. The stakes are high. "Is this right?" The humans collaborator knows 1 vs 2. The LLM doesn't know
-
LLM Security Challenges: Protection Against Emerging Attack Classes
By
–
Right, that's why the stuff is such a massive nightmare: so many of the things that we want to use LLMs for can't be done securely in the absence of a completely reliable protection against this class of attacks – which so far does not exist
-
Vulnerability mitigation reliability standards expectations
By
–
I think in some cases it is reasonable to expect 100% reliability from a vulnerability mitigation
-
CaMeL Approach Security: Preventing Untrusted Content in AI Prompts
By
–
You mean the CaMeL approach? Yeah actually implementing that well looks really difficult to me, you have to make about sure that any untrusted content is "tainted" and is then NEVER included in a prompt that might trigger tools