August 17, 2026
Today's top tech stories, deduped across the newsletters I read and briefly summarized. Click a source to open the original article.
AI
Anthropic publishes Claude's system prompts
Anthropic has made Claude's system prompts publicly available in its release notes, giving developers visibility into how the models are configured behind the scenes. The story drew heavy attention on Hacker News.
Are AI models being made dumber on purpose?
A blog post argues several AI vendors are quietly degrading model quality to cut costs without telling users. The claim sparked heavy discussion on Hacker News.
Stripe strikes $7 billion-plus deal to buy AI firm OpenRouter
Payments giant Stripe has reportedly agreed to acquire AI routing company OpenRouter for over $7 billion, according to Bloomberg. The deal underscores how much capital is now flowing into AI infrastructure.
The gray market for reselling AI credits is growing
An analysis looks at 'token brokers' who buy up cheap AI API credits and resell them for profit, a gray-market practice that undercuts vendors' pricing models.
Nvidia dramatically scales back OpenAI data center financing guarantee
Nvidia has significantly cut how much of a potential $250 billion data center deal with OpenAI it's willing to guarantee financing for, according to Reuters/WSJ. The pullback raises questions about risk in the massive AI infrastructure buildout.
Qwen 3.8 27B impresses but consistently overthinks
Alibaba's new Apache 2-licensed Qwen 3.8 27B is vision-capable and a practical size for a well-specced laptop, but tends to badly overthink simple tasks, according to Simon Willison.
Security
SafePal data breach hits nearly 40,000 customers
Cryptocurrency hardware wallet maker SafePal is warning of a data breach affecting about 39,798 customers after a flaw was exploited to steal order information. A threat actor now claims to be selling the stolen data.
Large-scale DDoS attacks disrupted secure messaging service Threema
Multiple large-scale distributed denial-of-service attacks hit the Threema secure messaging service earlier this week, causing severe communication disruptions.
New AmnesiaStealer macOS malware hijacks browser sessions
The new information-stealing malware AmnesiaStealer targets macOS users via ClickFix attacks and includes a streaming module that lets attackers interactively remote-control the victim's browser.
Dev
Protobuf now has LSP support
Buf has released LSP support for Protocol Buffers, giving developers autocomplete, go-to-definition, and other editor features directly in .proto files.
Web
Firefox for iOS gets a native ad blocker
Mozilla has added a native ad blocker to Firefox for iOS, letting users block ads without third-party extensions. The story was among the most popular on Hacker News this week.
IT
Anthropic confirms major Claude outage
Claude experienced a widespread outage, with users reporting login problems and degraded performance across several Anthropic services, according to BleepingComputer.