
Google's Threat Intelligence Group published evidence Monday of the first documented case of criminals using AI to discover and weaponize a zero-day exploit — a two-factor authentication bypass in a popular open-source admin tool. Google says it disrupted the campaign before it scaled. Cerebras raised its IPO range again ahead of pricing Wednesday night, with the offering now teed up at a basic-share market cap of $34.4B (fully diluted closer to $49B) and a book reportedly more than 20 times oversubscribed.
Google's Threat Intelligence Group published a finding Monday that researchers say is the first documented case of a criminal hacking group using an AI model to both discover a previously unknown software flaw andbuild a working exploit for it. The target was a two-factor authentication bypass in a widely deployed open-source web admin tool. Google says it worked with the vendor to quietly patch the issue before the campaign could trigger what researchers describe as a planned mass-exploitation event.
The tell that the exploit was AI-written is, almost comically, an LLM tell. The exploit code carries educational docstrings, a hallucinated CVSS severity score, and a textbook-Pythonic structure of the kind LLMs reproduce from their training data. The attackers, in other words, didn't bother to sand off the model's fingerprints.
John Hultquist, GTIG's chief analyst, framed it plainly: anyone treating AI-assisted vulnerability discovery as a future problem is already behind. "For every zero-day we can trace back to AI, there are probably many more out there."
For private-markets readers, the layer that matters is the read-across to the cybersecurity-defense cohort. The argument that AI-native cyber defense was a feature, not a category, has been running into trouble all year. Today it ran into a wall. The economics of CISO budgets are about to re-rate around an expected-this-year cost of a major AI-augmented breach, not a tail-risk one. Cybersecurity spending was already growing roughly 50% faster than general software, and is projected by Gartner to reach $240B globally in 2026. Public cyber multiples sit at a median ~9x NTM revenue against ~6x for broader SaaS — a roughly 25% premium — per Windsor Drake's Q1 2026 cybersecurity valuation report. The top-tier AI-and-cloud names (CrowdStrike, Cloudflare, Zscaler) trade at 14x–22x; that's the spread the secondary market has been waiting for a catalyst on.
The local proof points landed in the same news cycle. Frame Security, an Israeli startup founded this year by Wiz alum Tal Shlomo and former Team8 CTO Sharon Shmueli, emerged from stealth Monday with $50M, led by Index, Team8, and Picture Capital, with Wiz CEO Assaf Rappaport and Elad Gil participating. Google paid $32B for Wiz two months ago. The cohort that was being asked to justify its valuations is now sitting on a real-world demand signal.

Cerebras filed an updated S-1 Monday morning lifting its price range to $150–$160 per share from $115–$125, increasing the offering from 28 million to 30 million shares, and pricing now scheduled for Wednesday night. First trade Thursday on the Nasdaq under CBRS. At the top of the range, Bloomberg pegs the basic-share market cap at $34.4B; fully-diluted (counting options and the OpenAI MRA warrants), most coverage cites $48.8B. Reuters reports the book is more than 20 times oversubscribed. Morgan Stanley, Citi, Barclays, and UBS lead the syndicate.
The 24-day arc is what matters. The April 17 S-1 targeted $22–$25B. The May 4 amendment moved it to $26.6B. Monday's revision puts the fully-diluted mark at roughly twice that, with no new financial disclosures in between. The book changed. The company didn't.
For Pulse readers the relevant layer is the relationship to the secondary tape. Forge and Hiive desks were quoting Cerebras at $26–$28B through April. Pricing tomorrow night will land the IPO mark roughly 75% above where the same shares were clearing on anonymous secondary platforms three weeks ago. For most of this cycle, IPOs priced to or below the secondary mark — Klaviyo, Reddit, Instacart, Arm. Cerebras is the cleanest case in recent memory of the public bid pulling the offering decisively above the private mark. The greenshoe exercise in the first 30 days will be the cleanest tell of whether the book is genuinely broad or a thin cohort that had to own the largest IPO of 2026.

Nvidia's total committed AI-related equity investments year-to-date through May, per CNBC's reporting. That includes roughly $30B into OpenAI, $3.2B into Corning, $2.1B into IREN, and a long tail of smaller positions in CoreWeave and other AI-adjacent names. The takeaway worth carrying into a meeting today: the largest single financier of the AI capital cycle is now the company selling the picks-and-shovels. Nvidia is buying equity in the customers who buy its hardware — and the cohort of AI privates increasingly sits, at the cap-table level, downstream of one supplier.
A confidential draft registration statement — sometimes called a "confidential S-1" — is an option the SEC has given emerging growth companies (under $1.235B in revenue) since the 2012 JOBS Act, and extended to all issuers in 2017. The filer submits a full draft S-1 to the SEC, who reviews and provides comments, but the document remains nonpublic. The issuer can iterate with the SEC for months without exposing internal financials, competitive positioning, or risk-factor language to the market. The catch is timing: the company must publicly file the registration statement at least 15 business days before its roadshow begins. SpaceX confidentially filed April 1; its public S-1 window opens this Friday for a planned June roadshow. Robinhood Venture Fund II confidentially filed Monday for its second publicly traded venture vehicle — the same mechanism, applied to a registered investment company rather than an operating business.