<p><strong>TLDR:</strong> Cerebras priced its IPO at $185/share, raising $5.55B at a $56.4B valuation — the largest US tech IPO since Snowflake’s 2020 debut. OpenAI shipped Codex to iOS and Android, letting developers monitor and steer agentic coding sessions from their phones. An AI-assisted security firm found an 18-year-old critical RCE vulnerability in Nginx that needs patching now. OpenAI is also reportedly exploring legal action against Apple over the ChatGPT-Siri integration that failed to deliver expected revenue.</p>
<hr>
<h2>Cerebras IPO: $5.55B raised, $56.4B valuation</h2>
<p>Cerebras priced at $185/share on Nasdaq under ticker CBRS, nearly doubling its $26.6B April valuation in five weeks. A multi-year OpenAI contract for 750MW of inference capacity — worth over $10B at signing — was a key driver. The customer concentration risk is real: G42 and a related entity accounted for roughly 86% of 2025 revenue. Founder Andrew Feldman’s IPO roadshow pitch centered on three claims: inference demand will grow 1,000,000x, GPUs aren’t the only viable compute path, and CUDA’s moat is overstated — pointing to Gemini and Anthropic models trained without it.</p>
<h2>Codex goes mobile</h2>
<p>OpenAI launched Codex in preview on the ChatGPT mobile app for iOS and Android. The setup keeps files, credentials, and local environment on the host machine while streaming updates to the phone via a secure relay layer. You can review outputs, approve commands, switch models, and start new tasks remotely. Additional updates include Remote SSH, programmatic access tokens for Business and Enterprise plans, and HIPAA-compliant Codex for eligible Enterprise workspaces. Codex now has over 4 million weekly users. Anthropic launched a similar mobile monitoring feature for Claude Code back in February.</p>
<h2>18-year-old Nginx RCE, patch now</h2>
<p>Security startup DepthFirst AI used an LLM-powered platform to find four vulnerabilities in Nginx, including a critical heap buffer overflow (CVE-2026-42945, CVSS 9.2) in the URL rewrite module. The flaw affects Nginx 0.6.27 through 1.30.0 and Nginx Plus. On systems with ASLR disabled, Nginx’s multi-process architecture makes exploitation feasible through repeated attempts without memory layout changes. A public PoC is already on GitHub. Patches are in Nginx 1.31.0, 1.30.1, and Nginx Plus R36 P4 / R32 P6 / 37.0.0. If you’re running Nginx in production, this is the thing to do today.</p>
<h2>OpenAI vs. Apple</h2>
<p>OpenAI’s lawyers are exploring breach-of-contract options against Apple over the ChatGPT-Siri integration announced at WWDC 2024. OpenAI expected billions in annual revenue from the deal; Apple’s implementation buried ChatGPT behind explicit invocation requirements and constrained response windows. Meanwhile, Apple is moving on: iOS 27 introduces an Extensions system letting users pick from Claude, Gemini, and others, and Apple struck a roughly $1B/year deal with Google for Siri’s underlying AI. Any legal action is likely on hold until OpenAI’s ongoing federal trial with Elon Musk wraps up.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Recursive Superintelligence out of stealth:</strong> Richard Socher’s self-improving AI startup raised $650M at a $4.65B valuation with fewer than 30 employees and no shipped product; co-founders include ex-Meta FAIR director Yuandong Tian and Peter Norvig.</li>
<li><strong>Figma Q1 earnings:</strong> Revenue hit $333.4M, up 46% YoY; after enforcing AI credit limits on March 18, over 75% of users who exceeded free allocation kept paying, and net dollar retention reached 139% — a two-year high.</li>
<li><strong>SpaceXAI staff exodus:</strong> More than 50 researchers and engineers have left since February following SpaceX’s xAI acquisition, with the pre-training team shrinking to a handful after its lead departed.</li>
<li><strong>LangChain Labs launched:</strong> A new applied research effort from LangChain focused on continual learning for production agents, with Prime Intellect as a deployment partner.</li>
<li><strong>LangSmith LLM Gateway:</strong> A new runtime governance layer for agents offering cost limit enforcement, PII detection, and action controls.</li>
<li><strong>Anthropic + Gates Foundation:</strong> A $200M, four-year partnership targeting vaccine development for neglected diseases, K-12 tutoring tools, and agricultural productivity programs.</li>
<li><strong>Genkit middleware:</strong> Google’s open-source Genkit framework now ships composable middleware for TypeScript, Go, and Dart covering generate, model, and tool execution layers — with retries, fallbacks, approval gates, and scoped filesystem access out of the box.</li>
<li><strong>GPT-4.5 caching regression:</strong> At least one developer is seeing 2.5x input token usage compared to a week ago, suggesting prompt caching is degraded or broken on OpenAI’s end.</li>
<li><strong>GitHub Copilot app preview:</strong> A GitHub-native desktop app for agentic development that starts sessions from issues and PRs, works in isolated branches, and includes Agent Merge for auto-resolving review comments and failing checks.</li>
<li><strong>AI agent finds macOS kernel exploit:</strong> Security firm Calif used an AI system called Mythos Preview to develop the first public kernel memory corruption exploit targeting Apple M5 with Memory Integrity Enforcement enabled — built in roughly five days after initial bug discovery.</li>
<li><strong>AWS Security Agent:</strong> Full repository code review in preview uses a multi-agent pipeline to detect chained exploits and systemic vulnerabilities that pattern-matching SAST tools miss, at no extra charge to existing customers.</li>
<li><strong>PraisonAI CVE actively scanned:</strong> A critical auth bypass (CVE-2026-44338) in PraisonAI was scanned by attackers less than four hours after disclosure; authentication is disabled by default in affected versions 2.5.6–4.6.33. Patch to 4.6.34 immediately.</li>
<li><strong>Gemini 3.1 Flash TTS:</strong> Google launched a new TTS model scoring 1,211 Elo on the Artificial Analysis leaderboard, with audio tags for vocal style control, native multi-speaker dialogue, 70+ language support, and SynthID watermarking.</li>
<li><strong>Microsoft exploring Inception acquisition:</strong> Microsoft is reportedly in talks to acquire Inception, the Stanford-spun startup behind Mercury 2, described as the world’s first reasoning diffusion LLM — this follows the April 27 amendment ending Microsoft’s exclusive OpenAI licence.</li>
<li><strong>Conductor raises $22M Series A:</strong> The AI coding startup launched Conductor Cloud — persistent hosted workspaces where coding agents keep running after developers disconnect, following similar moves from Anthropic and Mistral.</li>
<li><strong>DORA AI ROI report:</strong> Google Cloud’s DORA team published findings that AI value depends more on organizational systems than the technology itself, follows a J-curve for value realization, and requires workforce retention and process redesign for sustained gains — though at least one analyst called it more marketing brochure than rigorous research.</li>
<li><strong>Cerebras CEO on inference speed:</strong> Andrew Feldman’s analogy is blunt — nobody would pay to downgrade from broadband to dial-up, and slow AI inference will become equally unacceptable once fast alternatives are ubiquitous.</li>
<li><strong>Welcome to the strip mining era of open source security:</strong> LLM-powered scanners are uncovering OSS vulnerabilities at roughly 10x historical rates; Metabase went from ~10 security reports per month to ~10 per week starting early 2026.</li>
</ul>


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

Agentic Digest

A packed week in AI and developer tooling: Cerebras priced its IPO at $185/share, raising $5.55B at a $56.4B valuation — the largest US tech IPO since Snowflake in 2020, driven largely by a $10B+ OpenAI inference contract. OpenAI shipped Codex to iOS and Android, enabling remote monitoring and steering of agentic coding sessions with 4M+ weekly users. A critical 18-year-old RCE vulnerability (CVSS 9.2) was found in Nginx via LLM-powered security tooling — patches are available and urgent. OpenAI is exploring legal action against Apple over the underperforming ChatGPT-Siri integration. Other highlights include GitHub Copilot's desktop app preview, LangChain Labs launch, LangSmith's LLM Gateway, an AI agent developing a macOS kernel exploit, AWS Security Agent preview, and a surge in LLM-powered OSS vulnerability discovery.

Cerebras prices the biggest US tech IPO since Snowflake, Codex lands on mobile