Retter Reboot #16

The Retter Reboot : Your weekly dose of ai & tech news

Anthropic Ships Claude Opus 4.7 — State-of-the-Art Coding With Built-In Cyber Safeguards

Anthropic released Claude Opus 4.7 on April 16, generally available across all Claude products, the API, Bedrock, Vertex AI, and Microsoft Foundry. The model is a notable step up from 4.6 on the hardest software engineering tasks, with better vision and more tasteful outputs for slides, interfaces, and docs. It's also the first model Anthropic has shipped with automatic cyber-request detection and blocking — a safeguard system being tested here before any broader release of Mythos-class models. Pricing unchanged at $5/M input, $25/M output.

Source: Anthropic

OpenAI Launches GPT-5.4 Cyber to Rival Claude Mythos — Different Philosophy, Same Fight

Just days after Anthropic's Project Glasswing reveal, OpenAI launched GPT-5.4-Cyber, a fine-tuned variant with a lower refusal boundary for legitimate security work. It adds binary reverse engineering and is rolling out via OpenAI's Trusted Access for Cyber program — a more scalable, verification-based approach versus Anthropic's invite-only partner model. Two philosophies emerging: Anthropic keeps the most powerful cyber model in a small coalition, OpenAI wants to scale access outward to more defenders with identity checks.

Source: NewsBytes

Anthropic's Mythos Sparks Emergency Wall Street Meeting — Treasury and Fed Summon Bank CEOs

Treasury Secretary Scott Bessent and Fed Chair Jerome Powell called an emergency meeting with CEOs of Citigroup, Bank of America, Morgan Stanley, Wells Fargo, and Goldman Sachs to discuss cyber risks from Anthropic's Mythos model. JPMorgan is already a Project Glasswing partner. The Treasury is now seeking its own access to Mythos to hunt for vulnerabilities in critical financial infrastructure. A striking signal: the US government is treating a private AI model as systemic infrastructure risk — and as a national security asset worth negotiating access to, even while the Pentagon has Anthropic blacklisted.

Source: Yahoo Finance

Google Adds Side-by-Side Browsing to AI Mode in Chrome — AI Becomes a Persistent Research Layer

Google rolled out a major AI Mode update in Chrome: clicking a link now opens the page alongside the AI conversation instead of replacing it. A new "+" menu lets users pull open tabs, PDFs, and images into a query as context. Strategic signal: Google is trying to turn Search into a persistent browsing partner rather than a closed answer box — partially addressing the publisher complaint that AI summaries bypass source sites entirely.

Source: TechCrunch

Microsoft Is Building Yet Another OpenClaw-Style Agent

Microsoft confirmed to The Information that it's testing OpenClaw-like features inside Microsoft 365 Copilot, aimed at enterprise customers with stronger security controls than the open-source runtime. The pitch: a version of 365 Copilot that's always working and can complete multi-step tasks over long periods. It joins Copilot Cowork (now Claude-powered) and Copilot Tasks in a growing agentic stack. Expected to debut at Microsoft Build in June. Meanwhile, Mac Minis continue selling like hotcakes as the unofficial OpenClaw runtime of choice.

Source: TechCrunch

Anthropic Plots a Lovable Challenger — Vibe-Coding Is the Next Battleground

Leaked screenshots inside Claude suggest Anthropic is building an in-chat app builder that generates AI chatbots, photo albums, and landing pages from simple prompts — putting it in direct competition with Swedish darling Lovable ($6.6B valuation in December). Lovable's head of growth recently said she worries more about "the big boys" — OpenAIs, Anthropics, Googles — than other vibe-coding startups. She was right. Expect the vertical AI stack to keep eating the vibe-coding layer.

Source: Sifted

Sequoia Raises $7B to Double Down on AI — First Mega-Fund Under New Leadership

Sequoia Capital closed roughly $7B for its expansion fund — nearly double the $3.4B raised in 2022 — the first major capital raise under new co-stewards Alfred Lin and Pat Grady after Roelof Botha stepped down. The fund fuels late-stage AI bets on both the foundation model giants (OpenAI, Anthropic — both eyeing 2026 IPOs) and application-layer companies like Physical Intelligence and Factory. Clear signal: late-stage AI investing has become a different game, and the firm with a 54-year pedigree is sizing up for it.

Source: TechCrunch

Meta Is Building an AI Clone of Mark Zuckerberg to Talk to Employees

Per the Financial Times, Meta is building a photorealistic, real-time AI Zuckerberg — trained on his mannerisms, tone, public statements, and strategic thinking — so its 79,000 employees can "feel more connected" to the CEO. Zuck is reportedly hands-on with training and testing. This is separate from his personal "CEO agent" that fetches company info for him. Choose your own subhead: genuinely useful internal tool, dystopia preview, or hiring freeze tell.

Source: Ars Technica

Apr 17, 2026