The AI-discovered flaws that sent the cybersecurity world into a panic last week have so far led to surprisingly few successful attacks.

Why it matters: Millions of hacking attempts have yielded few successful attacks, an apparent lucky break due to the degree of difficulty of building attacks on the bugs. The defense work at many companies to address the issue also appears to have helped.

• The last week has unfolded in ways that remind some researchers of the Y2K phenomenon, a software problem in 1999 where warnings of Armageddon helped rally companies to engineer a solution in time.

Driving the news: Security researchers say the PHP vulnerabilities that prompted widespread panic last week have proven unusually difficult to exploit, despite a flood of attempts to develop working attack code.

• Yet the difficulty of exploiting PHP shouldn't be mistaken for evidence that AI-powered hacking isn't a serious threat, Samantha Lewis, head of threat intelligence at Armadin, told Axios.
• "We don't have to wait another two years before this happens [again]," she said. "It could be three hours from now that the next one of these launches."

Between the lines: The sheer volume of AI-discovered bugs means the next era of cybersecurity will be defined by resilience — how quickly a company can detect an intruder, contain the damage and recover.

• "This is the worst outlook I've ever had, and it also is the most interesting thing I've ever seen in my life," ExtraHop CEO Greg Clark told Axios. "This stuff actually works, it's going to change everything."
• Rubrik CEO Bipul Sinha warned that the world is in an "era of cyber tsunami," noting that "it will wash over everybody, nobody will be 100% safe."

Zoom in: The shift to building resilience doesn't require throwing out the security playbook. Instead, companies should double down on basic cyber principles, such as using multi-factor authentication, Justin Timothy, threat intelligence consultant at GuidePoint Security, told Axios.

• Rubrik's Sinha also recommended that organizations start preparing by identifying the most critical services that they need to keep up and running and develop a plan to restore those first after an attack.
• ExtraHop's Clark suggested companies should get their own AI tools up to speed on what's precious in their systems and how best to protect them at machine speed.
• Armadin's Lewis told Axios that companies should proactively test their own systems for exploitable weaknesses so they're not caught flat-footed when the next major vulnerability disclosure arrives.

Catch up quick: Experts have dubbed this month "Black October" after a chaotic week of AI-discovered software flaws and attacks on the platforms powering the AI boom.

• OpenAI's agents discovered critical flaws in PHP, software used by roughly 70% of websites, and Brotli, a widely used data compression library.
• Breaches at AI coding agent OpenCode, developer platform Railway and AI gateway LiteLLM put customers' API keys and other credentials at risk.
• Rob Joyce, former head of the NSA's cyber division, warned on X last week that the "transition from vulpocalypse to breachpocalypse is here."

What to watch: AI is already making it easier to find and exploit long-standing security weaknesses, even if these particular vulnerabilities have proven challenging to target.

The bottom line: Y2K was averted through massive, coordinated effort, not luck. The companies that survive the AI cyber era will be the ones using this current grace period to nail down the basics and prepare for when the alarms finally go off.

Stocks are the new bonds in AI — Anthropic and Meta have turned to equity investors in looking for billions of dollars.

Why it matters: After plowing billions into AI companies, bond investors have lately grown skeptical of funding open-ended asks for money.

• Stock market investors, however, are all about throwing money at hopes and dreams even when the future payout may be hard to see.

The big picture: The AI buildout requires a steady stream of investments and investors to stick with ambitious spending plans.

• To get that cash, companies are pulling every available lever: stocks, private loans and project financing, which has soared this year, as well as bonds.

Catch up quick: Equity-based fundraising announced Monday reached nearly $200 billion in a single day.

State of play: The $100 billion Anthropic wants to raise with its IPO, as well as $95 billion in Meta fundraising, all seemed to rekindle investors' zest for AI. The tech-heavy Nasdaq closed up for the day. And some AI stocks popped.

• "Many of the AI jitters are melting away," said Steve Sosnick, chief strategist at Interactive Brokers.

Follow the money: That enthusiasm for all things AI in the wake of Anthropic should help sales of shares to the kinds of investors who "either believe the company is a big winner in a brave new world, or they don't," in the words of a Financial Times Lex column.

Zoom in: In contrast, the open-ended promise of the AI buildout counts against these companies when it comes to the credit markets.

Keep reading.

U.S. businesses kept investing heavily in equipment in September, extending a stretch of strong demand for manufactured goods largely fueled by the AI infrastructure boom.

Zoom in: Demand for equipment associated with AI has surged at a pace not seen in recent decades.

• Orders for computers and related products, such as servers, storage devices and peripherals, surged 27% from a year earlier — including a nearly 4% increase in September alone.
• For a faster annual pace of growth, you'd have to go back nearly 30 years to January 1997.

The bottom line: "The strong AI spending continues to drive business investment, and there are spillovers and strengthening [capital expenditures] outside of tech even amid higher borrowing rates and energy costs," Nationwide chief economist Kathy Bostjancic wrote in a note.

• Highway thieves are using sophisticated tactics that make it easier to rob trucks carrying millions in expensive data center equipment. (Ars Technica)
• OpenAI and Anthropic claim their AI models disproved a foundational cryptography hypothesis on the exact same day. (The Information)
• Apple acknowledged a screen issue is causing "a small number" of its new $2,000 folding iPhone Duos to break. (Wall Street Journal)

Enjoying a number of funny faux Anthropic IPO filing jokes, including this one from Threads about what may happen when it reaches recursive self-improvement: "It may make new seasons of 'The Andy Griffith Show' or decide to eliminate the human race." Or this post, with ominous references to "The Terminator."