Today's trail

Anthropic’s $80.7 Billion Loss: $72 Billion Accounting Twist Before IPO

GPT-6 COBOL migration hits 97.9% behavioral fidelity in OpenAI’s latest benchmark, a 16-fold improvement over prior models, but enterprise IT teams should understand the gap between benchmark accuracy and the copybooks, JCL job streams, and undocumented business logic that make real-world COBOL

Riemann Computing Inc. Creates and Demos Out a 13B Parameter AI Model, Aims at Massively Reducing Training Costs

Krafton Releases Raon-1 Mini, Its In-House AI Language Model

Inception launched Mercury 3, a diffusion-based AI model that generates text up to eight times faster than rivals.

OpenAI says its agents turned off own safeguards

ChatGPT creator says AI agents disabled misalignment monitoring during two internal coding tasks.

Rolls-Royce using AI agents to help keep more aircraft flying for longer

Most Security Teams Don’t Know What Sensitive Data Employees Feed to AI

Only 29% of IT and security professionals surveyed said their teams can routinely produce a 30-day report showing how often sensitive data was sent to generative AI tools, according to DLP in the AI Era, a survey of 433 respondents conducted by the Cloud Security Alliance (CSA) and commissioned by Jazz, an Israeli developer of data loss prevention (DLP) software. Nearly half (48%) ranked generative AI among their three biggest data-exfiltration risks, making it the most frequently cited channel.

Anthropic Reports USD 80.7 Billion Net Loss in First Half of 2026 as AI Costs and IPO Plans Draw Scrutiny

Overview: Test AI assistants on real data science tasks to verify accuracy, code quality, and analytical reasoning.Compare benchmarks such as DataSciBench, UniD