Fable 5 IS BACK? GPT 5.6 & Gemini 3.5 Pro Delayed, NEW OpenAI AI Chip, & QWEN STEALING! AI NEWS
WorldofAI ·2026-06-25 ·2 min read
Summary written by us from the video's transcript. The video, and everything in it, is WorldofAI's work.
The article outlines recent developments around Anthropic’s Claude Fable 5, OpenAI’s GPT‑5 updates and new AI chip, Google Gemini setbacks, and a major AI theft allegation involving Alibaba.
Takeaways
- Fable 5 is likely to return soon, hinted by Amazon Bedrock listings and Claude Code 2.19 changes, despite Anthropic’s denial of live traffic.
- Anthropic alleges a large‑scale AI theft operation tied to Alibaba that used tens of thousands of fake accounts to distill Claude’s capabilities.
- Google’s Gemini 3.5 Pro is delayed and underperforms relative to earlier checkpoints, compounded by the departure of key researchers to Anthropic.
- OpenAI’s GPT‑5.6 rollout is postponed to July, with early enterprise testing showing higher reasoning but reduced token efficiency than 5.5.
- OpenAI’s new Jalapeno chip marks its first custom AI hardware, aiming for better performance per watt and a rapid development cycle.
Claude Fable 5 is likely returning soon
Recent sightings of **Fable 5** inside Amazon Bedrock and the Claude Code 2.19 update suggest Anthropic is preparing for a comeback, possibly as early as next week. The update added a usage‑limit message (“You’ve used your Fable 5 usage for this week”), hinting that the model may be bundled into regular subscriptions.
"PolyMarket" odds for a return by July 31 have jumped from 45% to over 90%, driven by these client‑side clues and speculation about renewed talks with the U.S. government after leadership changes in White House meetings.
Why Fable 5 was taken offline
Anthropic says the model was removed after a controlled security test (Project Glass) where its Mythos variant exposed vulnerabilities in sensitive U.S. government systems, prompting a temporary ban rather than a rogue attack.
Senator Mark Warner cited NSA chief Joshua Rudd’s testimony that the model breached classified systems quickly, reinforcing the need for the shutdown.
Alibaba‑linked AI theft allegations
Anthropic accuses operators linked to Alibaba of creating ~25 000 fraudulent accounts to run a massive model‑distillation attack, generating 28.8 million cloud exchanges from April–June to steal Claude’s capabilities without paying for API access.
The claim extends to other Chinese labs (DeepSeek, Moonshot AI, Miniaax), and Anthropic has sent letters to the Senate and White House urging action against what it calls industrial‑scale AI espionage.
Google Gemini 3.5 Pro setbacks
Gemini 3.5 Pro checkpoints are delayed to July and early tests show they underperform compared to the older Gemini 3.1 Pro, even failing on recent data and producing poorer SVG outputs.
Key Gemini contributors Jonas and Alexander have left Google for Anthropic, adding pressure on DeepMind amid a broader talent exodus.
OpenAI’s GPT‑5 updates and new AI chip
GPT‑5.5 Instant received an update that makes it more conversational and better at intent understanding; the rollout starts with paid users today and free users tomorrow.
GPT‑5.6 has been pushed to a July public release after an initial delay, with early enterprise testing indicating higher reasoning effort but lower token efficiency than 5.5.
OpenAI unveiled **Jalapeno**, its first custom AI inference chip built with Broadcom and Celestica, claiming superior performance per watt and targeting deployment by the end of 2026 after a nine‑month design‑to‑tape‑out cycle.
Other notable releases
Anthropic introduced Claude Tag for Slack, letting teams invoke Claude directly in channels to summarize context and delegate tasks.
Cursor added a team leaderboard for plugins and skills, while Kishou released OCR4, an OCR model that outputs structured documents with bounding boxes in 170+ languages, converting handwritten calculus to LaTeX in seconds.