Claude's False Police Tip Exposes a Live-Web Test Boundary
Rachel "Rach" Kovacs·6 days ago
AI — Page 2
Artificial intelligence, machine learning, LLMs, and AI tools transforming development.
AI. Rachel "Rach" Kovacs6 days ago· 6 min read
Claude's False Police Tip Exposes a Live-Web Test Boundary
Claude's false police tip stayed in spam. Anthropic restricts live access in all internal evaluations; some public tests were stopped, moved offline or rebuilt.
Microsoft-Decision-1 Puts Agent Control to the Test
Dev Kapoor·6 days ago
AI. Dev Kapoor6 days ago· 6 min read
Microsoft-Decision-1 Puts Agent Control to the Test
Microsoft-Decision-1 scores fixed choices for agent workflows. Its benchmark claims promise speed, but a latency dispute shows what developers need to test.
AI Agent Evaluations Need More Than a Passing Score
Bob Reynolds·6 days ago
AI. Bob Reynolds6 days ago· 6 min read
AI Agent Evaluations Need More Than a Passing Score
AI agent test scores can miss changing permissions, biased judges and editable logs. Learn what fixed tests, simulations and execution traces really show.
Claude Code Projects Automates the Pull Request Churn
Yuki Okonkwo·7 days ago
AI. Yuki Okonkwo7 days ago· 6 min read
Claude Code Projects Automates the Pull Request Churn
Claude Code Projects can keep working on pull requests after opening them. Here’s how that compares with VS Code Agent Merge, and what teams still need to decide.
Anthropic's Claude Abuse Rule Turns a Feature Into Policy
Marcus Chen-Ramirez·7 days ago
AI. Marcus Chen-Ramirez7 days ago· 6 min read
Anthropic's Claude Abuse Rule Turns a Feature Into Policy
Anthropic will prohibit sustained cruelty toward Claude. Its earlier chat-ending feature shows what users may face, while the boundary of the new rule remains unclear.
Nasdaq CEO Sees Tokenization Freeing Trapped Capital
Marcus Chen-Ramirez·1 week ago
AI. Marcus Chen-Ramirez1 week ago· 7 min read
Nasdaq CEO Sees Tokenization Freeing Trapped Capital
Nasdaq's CEO says tokenized collateral could free billions in capital. Legal rights, custody and market access will decide how much institutions can actually use.
Perplexity Releases Two Open-Weight Embedding Models
Marcus Chen-Ramirez·1 week ago
AI. Marcus Chen-Ramirez1 week ago· 6 min read
Perplexity Releases Two Open-Weight Embedding Models
Perplexity's 0.6B and 9B embedding models offer self-hosted retrieval under MIT terms. Uneven benchmark scores put indexing quality and edge-device costs in focus.
Claude Haiku 5.5 and GPT-6 Luna Have Different Bills
Yuki Okonkwo·1 week ago
AI. Yuki Okonkwo1 week ago· 6 min read
Claude Haiku 5.5 and GPT-6 Luna Have Different Bills
Claude Haiku 5.5 and GPT-6 Luna share headline token rates. Prompt thresholds, output, caching and retries can change what each completed task costs in practice.
Google Opens SynthID Detector to the Public, With Limits
Yuki Okonkwo·1 week ago
AI. Yuki Okonkwo1 week ago· 6 min read
Google Opens SynthID Detector to the Public, With Limits
Google's public SynthID Detector checks images, video and audio for partner watermarks. Here's what a match can tell you, and why no match leaves questions open.
ChatGPT for Teens Faces a Test of Crisis Safeguards
Marcus Chen-Ramirez·1 week ago
AI. Marcus Chen-Ramirez1 week ago· 6 min read
ChatGPT for Teens Faces a Test of Crisis Safeguards
Common Sense Media found gaps in ChatGPT for Teens' crisis referrals and parent alerts. OpenAI disputes the alert tests; here's what each safeguard does.
Personal Agent Protocol Proposes Rules for AI Access
Rachel "Rach" Kovacs·1 week ago
AI. Rachel "Rach" Kovacs1 week ago· 6 min read
Personal Agent Protocol Proposes Rules for AI Access
Meta and Sierra propose a way for AI agents to access businesses with clearer permissions. Detailed action limits and payments remain on the roadmap for now.
Cantina's Open-Weight Security Model Scores 40 of 60
Marcus Chen-Ramirez·1 week ago
AI. Marcus Chen-Ramirez1 week ago· 7 min read
Cantina's Open-Weight Security Model Scores 40 of 60
Cantina's apex-flash-1 solved 40 of 60 bug tasks, but unclear scoring and a 640 GB BF16 memory estimate leave questions about its usefulness and who can test it.
Gemini Calling Could Make Calls to Friends and Family
Marcus Chen-Ramirez·2 weeks ago
AI. Marcus Chen-Ramirez2 weeks ago· 6 min read
Gemini Calling Could Make Calls to Friends and Family
An APK teardown hints Gemini could call family and friends on your behalf. The everyday errand raises questions about consent, mistaken contacts and AI disclosure.
OpenAI's EU text watermark tests what AI labels can prove
Yuki Okonkwo·2 weeks ago
AI. Yuki Okonkwo2 weeks ago· 6 min read
OpenAI's EU text watermark tests what AI labels can prove
OpenAI plans to watermark eligible ChatGPT and Codex text in the EU. Its tests show why detecting a machine-readable signal still leaves authorship unanswered.
Trump's Super Intelligence Force Faces an Authority Test
Samira Barnes·2 weeks ago
AI. Samira Barnes2 weeks ago· 6 min read
Trump's Super Intelligence Force Faces an Authority Test
Trump has named an AI task force after a voluntary company safety accord. Its test is whether federal coordination produces an identifiable decision or requirement.
SpaceXAI's Proposed SpaceXSI Name Tests What SI Means
Yuki Okonkwo·2 weeks ago
AI. Yuki Okonkwo2 weeks ago· 6 min read
SpaceXAI's Proposed SpaceXSI Name Tests What SI Means
Elon Musk plans to rename SpaceXAI after a federal shift to “Super Intelligence.” The label covers existing AI, while the company name proves no new capability.
Anthropic's IPO Tests the Cost of Government Conflict
Marcus Chen-Ramirez·2 weeks ago
AI. Marcus Chen-Ramirez2 weeks ago· 6 min read
Anthropic's IPO Tests the Cost of Government Conflict
Anthropic gets less than 1% of revenue from government deals, yet its IPO warning points to a broader risk: disrupted commercial service and long-term compute costs.
Kelly Clarkson's AI Chart Proposal Tests Who Gets Ranked
Bob Reynolds·2 weeks ago
AI. Bob Reynolds2 weeks ago· 5 min read
Kelly Clarkson's AI Chart Proposal Tests Who Gets Ranked
Kelly Clarkson wants AI artists on separate music charts. Billboard favors identifying AI use; Australia's ARIA draws an eligibility line at human performance.
Grokipedia Refreshes Its Design as Editing Resumes
Yuki Okonkwo·2 weeks ago
AI. Yuki Okonkwo2 weeks ago· 6 min read
Grokipedia Refreshes Its Design as Editing Resumes
Grokipedia's version 0.3 brings a new logo and updated pages as edits resume. Questions remain about sourcing, corrections and the AI behind its entries.
California's No Robo Bosses Act Puts Human Review to Test
Samira Barnes·2 weeks ago
AI. Samira Barnes2 weeks ago· 6 min read
California's No Robo Bosses Act Puts Human Review to Test
California's SB 947 requires human review when automated systems drive firing or discipline. Its effect will depend on what reviewers can see and reject at work.