AI-powered cyberattacks fuel 'Black October' fears as stolen API keys rack up hundreds of dollars in unwanted charges

Hackers' growing use of artificial intelligence to automate attacks has produced a troubling wave of cybersecurity developments in October 2026. Some cybersecurity professionals have started calling the turbulent period ‘Black October’. The term describes a series of cybersecurity emergencies rather than a single coordinated hacking campaign.

One of the clearest personal examples is startup founder Colin Parsons, who woke up to roughly 50 billing notifications he never expected. He reported $281 in unexpected OpenAI usage and another $168 involving Anthropic — $449 in charges he did not recognise. Both credentials were associated with OpenCode, an AI coding platform he used.

On 17 October, OpenCode disclosed a security incident involving its Console service. The company said unauthorised access to a legacy system may have exposed third-party AI provider keys stored by some users. It reportedly notified affected customers and advised them to revoke their credentials.

OpenCode said the incident affected approximately 1.4% of its Console users who had connected external AI providers. The disclosure concerned credentials stored in Console, rather than keys stored locally in OpenCode's desktop or command-line applications. Other users also reported unexpected charges, although the amounts have not been independently verified. OpenCode associated its incident with a large-scale AI model distillation attack, but has not publicly established exactly how the reported unauthorised usage occurred.

The concern is straightforward: if someone obtains a working API key, they may be able to use a paid AI service and generate charges against the account owner's billing arrangement. For people who depend on AI tools for work, that can become expensive before they realise anything has happened.

Another major example comes from Microsoft. On 21 October, Microsoft Threat Intelligence revealed that a China-state-aligned hacking group called Storm-0940 had used an AI-powered penetration testing framework called ARTEX. The open-source tool can automate security testing tasks, including investigating networks and identifying potential weaknesses; in the hands of criminals, those capabilities can accelerate attempts to gain unauthorised access.

According to Microsoft, the group used ARTEX to target hundreds of organisations across multiple industries. Several attacks resulted in unauthorised privileged access to cloud systems, and in multiple cases attackers reached that level of access in less than three hours. Although many attempts failed, the successful intrusions demonstrate how quickly AI-assisted attacks can unfold.

Between 10 and 14 October, Microsoft recorded more than 151,000 failed login attempts in one campaign targeting 327 accounts across eight organisations. At its peak on 11 October, attackers generated approximately 2,900 failed login attempts per minute. The technique, known as password spraying, involves trying passwords against multiple accounts to identify a working login.

Importantly, Microsoft did not observe successful logins during that specific campaign. However, it documented unauthorised access and suspicious activity in other environments associated with the same threat group. Microsoft also connected Storm-0940 to a compromise involving ConnectWise, a company that provides software used by IT service providers; attackers exploited access associated with ConnectWise cloud applications to attempt entry into customer environments, and some then carried out further activity through Microsoft cloud services, including changes that could help them maintain access. Microsoft cautioned that suspicious access does not automatically mean customer information was stolen. Nevertheless, the activity reveals how AI can help attackers move quickly once they obtain usable credentials.

While hackers are becoming more capable, AI-assisted security research is also uncovering dangerous weaknesses in software that supports the internet. On 21 October, developers released emergency fixes for two critical vulnerabilities in PHP, a programming language widely used to operate websites.

The vulnerabilities, identified as CVE-2026-109618 and CVE-2026-109619, are collectively known as PHP2Shell. Under certain conditions, the flaws could allow attackers to execute malicious code on vulnerable servers without first signing in. Successful exploitation could expose sensitive information or give criminals control over affected systems.

The potential impact is enormous because PHP powers roughly 70% of websites whose server-side programming language is known. However, that does not mean 70% of websites were hacked, nor that attackers could successfully compromise every website using PHP. The vulnerabilities affect PHP installations handling web requests, but successfully exploiting them can require complex technical conditions.

The situation prompted warnings from US cybersecurity authorities. On 23 October, the FBI issued a public alert urging organisations to patch affected systems immediately. The Cybersecurity and Infrastructure Security Agency also called for urgent updates. The PHP project released fixes in versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Website administrators running older unsupported versions need to obtain appropriate vendor-supported fixes or migrate to a supported release.

Meanwhile, several government organisations in Europe temporarily took websites offline as a precaution. Analysts say the biggest concern is speed: traditionally, hackers needed significant time and technical expertise to investigate targets, develop attack methods and analyse what they obtained; now AI tools can automate portions of that work, operate continuously and pursue multiple potential targets at once — leaving companies less time to respond when criminals discover a way into their systems.