Claude Code’s Sandbox Bug Took 50 Days to Fix
💜 Software · AI Coding Tools Claude Code’s Sandbox Bug Took 50 Days to Fix A stealth security startup flagged
Software
GPT-6 Astra, OpenAI’s Model Just Hit a 100% Exploit Score
GPT-6 Astra launched September 3, 2026 as OpenAI’s first Critical-threshold model, scoring 100% on ExploitBench and beating Anthropic’s Claude Opus 5.
Claudeforce, Salesforce’s 37-Skill Bet on Anthropic’s Claude
Claudeforce made Claude the default reasoning engine for Salesforce’s Agentforce, launching a plugin with 37 prebuilt sales skills on August 26, 2026.
Stripe Acquires OpenRouter for $7B, SpaceX Takes Cursor for $60B
Stripe OpenRouter $7B deal closed 90 days after $1.3B Series B. SpaceX paid $60B for Cursor. AI’s middle layer is now tech’s priciest real estate.
Meta Open Sources Muse Spark, Zuckerberg Says Distribute Superintelligence
Meta open source: Muse Spark 1.2 weights released, Muse Glimmer runs on laptops. Zuckerberg’s 6,500-word manifesto targets OpenAI and Anthropic by name.
AWS Hits 18 Quarter High at 37%, But Meta Just Burned $31B in One Quarter
AWS hits 37% growth, its fastest in 18 quarters. Meta’s $31B Q2 capex left just $784M in free cash flow. The Q2 Big Tech cloud scorecard is complete.
Microsoft Anthropic Bet Paid $3.2B, Beat OpenAI All Year
Microsoft Anthropic investment paid $3.2B in Q4 FY26 alone, nearly matching OpenAI’s full-year $5B gain. Azure passed $100B, Copilot hit 30M seats.
Google Cloud Doubled Azure Growth to 82%, Alphabet Stock Fell Anyway
Google Cloud grew 82% to $24.8B in Q2 2026, more than double Azure’s 40%. Backlog hit $514B. Yet Alphabet fell 3.65% on $205B capex hike. Here’s why.
Software Selloff Hits 25 Year Peak, Microsoft $400 Broken and CRM Down 35%
Software Selloff hits 25-year peak on July 22 as Microsoft fell below $400, Salesforce down 35% YTD, S&P Software Index off 20%. Wedbush calls it historic.
The AI Paradox, GitLab Says 79% Code Faster But Nothing Ships Faster
AI Paradox: GitLab’s 1,528-dev survey finds 79% code faster but delivery flat. 92% report governance gaps, 43% can’t tell AI code from human.