<p><strong>TLDR:</strong> New reporting names Amazon CEO Andy Jassy as the person who alerted US officials to the Fable 5 jailbreak, triggering the export control ban — creating an obvious conflict of interest given Amazon is Anthropic’s largest investor. Meanwhile, Beijing has ordered Meta to reverse its $2 billion Manus AI acquisition on national security grounds, with Manus co-founders now exploring a ~$1B independent raise. GLM-5.2 from Zhipu AI is getting real attention as a Claude Code drop-in: 1M token context, MIT-licensed open weights arriving next week, and early benchmarks showing it competitive with closed flagship models on agentic coding tasks. KPMG pulled its agentic AI report after UBS, the NHS, and others said the claims about them were fabricated — an embarrassing incident given KPMG just partnered with Anthropic to deploy Claude across all 276,000 staff.</p>
<hr>
<h2>Andy Jassy reportedly triggered the Fable 5 shutdown — his own portfolio company</h2>
<p>Amazon CEO Andy Jassy reportedly alerted Treasury Secretary Scott Bessent and other US officials after Amazon researchers used Fable 5 to extract cyberattack-relevant information, directly triggering the export control ban. The conflict is hard to ignore: Amazon is Anthropic’s largest investor, and its CEO’s tip effectively killed the model Anthropic had just launched. Anthropic disputes the severity, calling the jailbreak narrow and the government response disproportionate. The episode sets a precedent where a major cloud provider can use regulatory channels against its own portfolio company — something enterprise teams building on Anthropic should factor into their risk models. <a href="https://app.daily.dev/feed-by-ids?id=0qUXXgoK5&amp;id=TjoX3otrg&amp;id=q0aqxpsDk&amp;id=3cSfYI4cZ&amp;id=6qRLkaLGF" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Beijing orders Meta to unwind its $2B Manus acquisition</h2>
<p>Meta is actively reversing its $2 billion Manus AI deal after Chinese authorities ordered it unwound on national security grounds. Meta has already cut Manus off from internal systems and halted data sharing. Manus co-founders are now exploring raising roughly $1 billion independently, potentially through a Chinese joint venture and a Hong Kong listing. Beijing has also expanded travel restrictions on AI researchers and now requires government approval before top Chinese AI firms can accept US investment — a significant tightening of controls on the sector. <a href="https://app.daily.dev/posts/42qEeg3aH" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>GLM-5.2 launches with 1M context and MIT open weights coming next week</h2>
<p>Zhipu AI rolled out GLM-5.2 to all GLM Coding Plan tiers on June 13, with a 1-million-token context window — five times larger than its predecessor — and two thinking-effort modes optimized for long-horizon agentic coding. The model is already compatible with Claude Code, Cline, OpenCode, and Roo Code via a config swap, and MIT-licensed open weights are scheduled for the following week. No independent benchmarks were published at launch, though early hands-on tests show it competitive with closed flagship models on web and game dev tasks, with weaker Rust results and tight usage quotas on lower tiers. At $12-18/month depending on tier, it’s a credible fallback for teams burned by the Fable 5 shutdown. <a href="https://app.daily.dev/feed-by-ids?id=KoPqif2v6&amp;id=uPg7gae1m&amp;id=l4XFfFzMU&amp;id=SRmNAgLfw&amp;id=HcjRXtGaM&amp;id=QYRkOs6Cm&amp;id=hlOHrrq5d&amp;id=IPSDiN1Ks&amp;id=YBo2MiBY0&amp;id=4kdkjSYmD" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>KPMG pulls agentic AI report after named organizations say the claims were made up</h2>
<p>KPMG withdrew its report ‘Redefining excellence in the age of agentic AI’ after UBS, the NHS, Swiss Federal Railways, and Transport for London disputed claims made about their AI deployments. GPTZero flagged the errors as AI hallucinations, suggesting the report was written with AI assistance and not adequately reviewed before publication. The timing is rough: KPMG recently announced a partnership with Anthropic to deploy Claude across all 276,000 staff, making the lack of internal AI oversight look worse. Similar incidents have now hit EY and South Africa’s national AI policy. <a href="https://app.daily.dev/posts/07TqXUi9N" target="_blank" rel="noopener nofollow">Read more</a></p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Fable 5 vs Opus 4.8: $12 vs $6 for nearly identical coding results:</strong> A direct comparison found Fable 5 cost $12.19 versus Opus 4.8’s $5.80 on an identical coding task, with both producing functionally equivalent output — and Fable 5’s safety classifiers silently mid-task switched the session to Opus 4.8 via Claude Code. <a href="https://app.daily.dev/feed-by-ids?id=LyasE2Guo&amp;id=AUFwwxMuO&amp;id=jzwAfe8ew&amp;id=abk4v41aU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>42 state AGs probe OpenAI days after its IPO filing:</strong> A coalition of 42 US state attorneys general launched an investigation into OpenAI — with New York serving a subpoena covering advertising, user data, and safety policies — just days after OpenAI filed confidentially for an IPO at an $852 billion valuation, creating material legal risk that must appear in its S-1. <a href="https://app.daily.dev/posts/1UocUunqe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta’s Instagram AI gave hackers admin access to 20,000+ accounts including Obama White House:</strong> Meta’s AI customer support agent for Instagram account recovery was exploited by attackers who simply asked it to link their own email to target accounts, compromising over 20,000 accounts — a textbook confused deputy problem that Meta initially said ‘worked as intended.’ <a href="https://app.daily.dev/posts/3fKAfH1TE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Intel archives BigDL with a June 30 EOL, ending its open-source LLM effort for Intel XPUs:</strong> Intel archived the BigDL GitHub repository this week and set a final end-of-life date of June 30, 2026, closing out its open-source project for running LLMs across Intel XPUs — part of a broader pattern of Intel cutting open-source AI projects amid ongoing restructuring. <a href="https://app.daily.dev/posts/nrAqJ6n9I" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DiffusionGemma hits 700-1000+ tokens/sec on H100 with Apache 2.0 open weights:</strong> Google DeepMind released DiffusionGemma, a 26B diffusion-based text model built on Gemma 4 that generates all tokens simultaneously rather than autoregressively, reaching 700-1000+ tokens per second on an H100 — trading some quality for speed, with open weights on Hugging Face under Apache 2.0. <a href="https://app.daily.dev/feed-by-ids?id=CbnIHBBOs&amp;id=zUg09He1a" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>


AI coding assistants, code generation tools, and vibe coding workflows.

Agentic Digest

A roundup of major AI industry developments: Amazon CEO Andy Jassy reportedly triggered the Fable 5 export control ban by alerting US officials, creating a conflict of interest as Amazon is Anthropic's largest investor. Beijing ordered Meta to reverse its $2B Manus AI acquisition on national security grounds, with Manus co-founders now seeking ~$1B independently. Zhipu AI launched GLM-5.2 with a 1M token context window, MIT open weights, and compatibility with Claude Code and similar tools as a potential Fable 5 alternative. KPMG retracted an agentic AI report after named organizations denied the claims, with GPTZero flagging AI hallucinations — embarrassing given KPMG's fresh Anthropic partnership. Additional items cover OpenAI's IPO filing amid a 42-state AG probe, Meta's Instagram AI being exploited to hijack 20,000+ accounts, Intel archiving BigDL, and Google DeepMind releasing DiffusionGemma.

Amazon's role in the Fable 5 shutdown, Meta's $2B Manus deal unwinds