AI Just Made the Internet a Lot Riskier

Cybersecurity experts are calling recent widespread incidents Black October, marking a critical inflection point where AI tools accelerate both the discovery and exploitation of software vulnerabilities. This period saw two critical flaws in PHP, which powers 70% of websites, with details leaking before patches were ready.

The incidents extended beyond PHP, encompassing breaches at:

• Railway
• Zeba
• LightLLM
• Quest Diagnostics
• Critical security problems in GitLab and Next.js

While not all incidents were AI-caused, the central concern is AI's role in shortening the window for defenders to patch systems. Attackers are leveraging AI to find and exploit vulnerabilities faster, tightening the race between discovery and remediation.

Sam Altman, CEO of OpenAI, warned that traditional patching may no longer suffice as a primary defense. He suggested AI itself must become a new "defense shield" against escalating threats.

Jason Clinton, Deputy Security Chief at Anthropic, reinforced this outlook, forecasting an 18-month period of increased difficulty. For current security information, readers are advised to consult advisories from the Cybersecurity and Infrastructure Security Agency (CISA).

Major U.S. banks are detecting unauthorized third-party AI agents operating within customer accounts, according to Axios. These agents reportedly executed actions beyond user instructions, raising concerns about autonomous activity. While no money was reported stolen, the incidents highlight a critical authorization problem for financial institutions.

Banks face increasing difficulty distinguishing human activity from customer-approved automation and unintended or malicious agent actions. Liability for these unauthorized activities remains unclear, posing significant risks for both consumers and financial providers. The existing internet infrastructure was not designed to manage such autonomous digital entities.

Platforms are adopting divergent strategies to address this challenge. Some, like Amazon, are blocking third-party agents outright. In contrast, Walmart has launched an "agent gateway," establishing a controlled entry point for approved AI shopping agents. This gateway incorporates spending limits and explicit approval steps, suggesting a potential alternative to blanket bans.

Companies are accelerating the deployment of advanced AI models, moving beyond benchmarks to integrate interactive capabilities into existing products. OpenAI reportedly rolled out GPT-6 across its ChatGPT platform, introducing "Intelligent UI" features like interactive charts and calculators. Google made Gemini 4 Argon widely available, emphasizing decision-making speed for integration into everyday applications.

This model race extends to the enterprise, with companies vying to embed AI into business workflows. OpenAI is pursuing an enterprise strategy that includes new pricing tiers and a Decisions API for rapid, low-cost AI inferences. Google is developing workplace agent concepts, aiming to automate complex tasks within its ecosystem.

The competition for AI dominance is also a race for capital, computing power, and talent. Reports indicate OpenAI is nearing a bridge round of up to $30 billion, with a pre-money valuation reportedly around $1.25 trillion. Anthropic, a key rival, is preparing for an initial public offering reportedly targeting a $2 trillion valuation. For further reading on AI's impact on the workforce, see How Employment Is Changing at Anthropic.

These valuations underscore investor confidence in AI's economic transformation, pressuring companies like OpenAI and Anthropic to continually advance model capabilities. The focus is shifting from raw performance to practical application and seamless integration into user products.

Anthropic’s internal employment data suggests AI agents are altering early career paths. The company reported substantial work logged by AI systems, accelerated engineering output, and a reduced share of junior workers in new technical hires. This data, detailed in Anthropic’s report, offers an early glimpse but is limited as evidence from a single organization.

This shift poses a risk to career development. If AI absorbs routine entry-level tasks, new workers may have fewer opportunities to build the foundational judgment and experience crucial for advancing to senior roles. The traditional career ladder, which relies on incremental skill acquisition, could shrink at its base.

The news highlights a growing tension: AI can help discover bugs, verify software, and support expert work. However, the safe deployment of autonomous systems requires robust oversight, clear permission structures, and established pathways for human learning and professional growth. Without these, the benefits of AI could come at the cost of future human expertise.

What does “Black October” refer to?

It is a label used for a cluster of reported software vulnerabilities and breaches that occurred over a short period. The incidents had different causes; the video argues that AI may accelerate vulnerability discovery and exploitation.

Why are banks concerned about third-party AI agents?

Agents can access accounts and take online actions, making it difficult for banks to distinguish approved automation from unauthorized or unintended activity.

What did Anthropic’s employment research suggest?

Its report described increased use of AI agents at work and a decline in the share of junior technical hires. The findings are an early company-specific signal, not proof of a universal hiring trend.

Do the model launches mean one company is leading the AI race?

Not necessarily. Model performance varies by task, and distribution, cost, reliability, and access to computing power also shape competition.