AI News, 14 November 2026

Saturday 14 November

Anthropic says outdated prompting habits are slowing down Claude Opus 5.5

AI-generated summary Anthropic states that outdated prompting habits carried over from older model generations are slowing down Claude Opus 5.5. Anthropic advises users to drop phrases like think hard from prompts and to start at the Medium effort level.

Also covered byaiagentnews.ioGPLP
Earlier:

How Prompt Injection Detection Works: Key Steps For Enterprise AI Security Teams

5 Prompt Injection Detection Techniques Enterprise AI Security Teams Must Know

AI-generated summary BlackFog reports that enterprise artificial intelligence security teams face growing prompt injection threats as businesses build more operations around assistants and agents. BlackFog states that ADX Vision 2.0 provides endpoint prompt injection detection with seven layers of protection for both employee typed inputs and autonomous workflows.

LlamaIndex launches ParseBench-2 for severity-aware document parsing evaluation

AI-generated summary LlamaIndex has released ParseBench-2 to evaluate document parsing quality with a focus on complex table parsing and a new severity policy layer. The benchmark includes a taxonomy of over 330 error types organized into five severity levels to help teams customize their evaluation metrics.

Earlier:

FINAL-Bench Launches Daily AI Weather Leaderboard

FINAL-Bench Launches Daily AI Weather Leaderboard

AI-generated summary Research initiative FINAL-Bench launched the Global Weather Model Leaderboard on Hugging Face to benchmark artificial intelligence weather models and physics-based systems against daily atmospheric observations. The evaluation framework tracks operational forecasts across twenty-four reference cities on six continents, measuring meteorological variables such as temperature, wind speed, and precipitation.

AWS Launches Bedrock AgentCore V2 & Search API

AWS Launches Bedrock AgentCore V2 & Search API

AI-generated summary Amazon Web Services announced the general availability of the Amazon Bedrock AgentCore V2 runtime and a Web Search API. The release introduces Firecracker microVM snapshot restoration that reduces session cold starts to under two seconds.

Earlier:

Gemini made my Google Home worse at the basics, so I moved my voice control to Home Assistant

AI-generated summary Google transitioned its smart voice architecture from Google Assistant to its large language model, Gemini, to provide conversational context and natural language understanding. The author experienced execution degradation and custom routine failures in Google Home after the Gemini migration, prompting a move to Home Assistant for local control.

Earlier: