JupiteX Get the app
Defence & Security17 Nov 2026 · about 7 min

Attackers are using AI agents to automate cybercrime at 'industrial scale,' FBI and AWS warn

The brief

An AI agent is software that can pursue a goal by choosing and carrying out steps, often using tools such as search, calendars, or business systems. A basic chatbot mainly produces a response to a user’s prompt. The supplied article discusses AI computing and security headlines, but it does not explain AI agents. For example, a chatbot might suggest how to arrange a meeting. An agent could check a calendar, find an open time, and prepare an invitation, subject to its permissions. The key difference is taking actions across steps, not simply giving a longer or more polished answer. Some systems combine both styles. Agents are not automatically independent or reliable: people set their goals, connect their tools, and define what they may do. Their ability to act can make them useful, but it also means errors or overly broad permissions can have consequences. Organizations should test agents and limit access to what each task requires.

01

What is an AI agent, and how is it different from a chatbot that only answers prompts?

An AI agent is software that can pursue a goal by choosing and carrying out steps, often using tools such as search, calendars, or business systems. A basic chatbot mainly produces a response to a user’s prompt. The supplied article discusses AI computing and security headlines, but it does not explain AI agents.

For example, a chatbot might suggest how to arrange a meeting. An agent could check a calendar, find an open time, and prepare an invitation, subject to its permissions. The key difference is taking actions across steps, not simply giving a longer or more polished answer. Some systems combine both styles.

Agents are not automatically independent or reliable: people set their goals, connect their tools, and define what they may do. Their ability to act can make them useful, but it also means errors or overly broad permissions can have consequences. Organizations should test agents and limit access to what each task requires.

02

What does “industrial scale” mean for cybercrime, and how could the volume or speed of attacks be measured?

In cybercrime, “industrial scale” describes attacks organized for repeatability and large reach, rather than one-off attempts carried out by hand. The supplied article includes security headlines, but it does not discuss automated cybercrime, attack volumes, or measurement methods.

Scale can be described with several separate measures: attempts per hour or day, unique accounts or organizations targeted, and the share of attempts that succeed. Defenders might also track how quickly attacks spread or how many alerts they generate. A high attempt count does not by itself mean many victims; success rates and confirmed harm matter too.

These measurements usually come from system and security logs, abuse reports, and investigations, which may not capture every attempt. Comparisons are most useful when they cover the same period and define “attack” consistently. Reporting both attempted activity and verified impact helps distinguish a noisy campaign from one that caused substantial damage.

03

What kinds of cybercrime tasks can attackers automate with AI agents?

An AI agent could help automate parts of a cyberattack, but the supplied article does not describe that activity. At a high level, possible tasks include sorting public information, drafting or adapting deceptive messages, translating text, and organizing repetitive work. These examples describe risks, not instructions for carrying out attacks.

A criminal might try to use software to prepare many message variations or handle routine steps in a campaign. Other automation, not necessarily AI, can test exposed services or attempt logins at scale. The important mechanism is that software repeats tasks quickly and can pass results from one step to another, while a person may still set objectives or review outputs.

Automation does not guarantee success: systems can make mistakes, and defenses can block or flag activity. Organizations can reduce harm by watching for unusual patterns, limiting access, and responding quickly. The article’s security items concern phishing against Signal users and a SharePoint zero-day, but it does not connect either story to AI agents.

04

Who are the FBI and AWS, and what does each bring to a warning about cyber threats?

The FBI is the United States’ Federal Bureau of Investigation, a federal law-enforcement agency. AWS, or Amazon Web Services, is a cloud-computing provider. The supplied article mentions neither organization, so it does not describe a joint warning or say what either contributed to one.

In general, the FBI can contribute information from investigations and its work responding to crime. AWS can bring a provider’s view of cloud infrastructure and security issues affecting its services or customers. Those are different perspectives: one is an investigative agency, while the other operates technology services. A specific warning would need to be checked for its stated authors and evidence.

Together, such perspectives could help explain both criminal activity and risks to cloud systems, but that is general context, not a fact reported in this roundup. It would be inaccurate to attribute particular findings, threat figures, or advice to the FBI or AWS based on the supplied text. The article’s security headlines instead cover Signal phishing, SharePoint attacks, and other topics.

05

What can happen to people and organizations if cyberattacks become faster and reach more targets?

If attacks become faster and reach more targets, more people and organizations may face attempts to steal information, take over accounts, or disrupt services. The supplied article does not describe AI-driven attacks or document these outcomes; these are general cybersecurity risks, not reported events in the roundup.

A successful phishing attempt, for example, can lead someone to reveal credentials. If those credentials provide access to work systems, the harm can extend beyond one person. Organizations may have to investigate, restore services, notify affected people, and manage financial or operational losses. The actual consequences depend on what was accessed and how quickly the incident is contained.

More attempts do not mean every attempt succeeds. Strong authentication, monitoring, and well-practiced response plans can reduce the chance that an attack causes lasting harm. The article does mention Russian actors posing as Signal support to launch phishing attacks and a zero-day attack on on-premises SharePoint, but it gives no victim counts or losses for either headline.

06

What safeguards can help people and organizations detect or limit automated cyberattacks?

Organizations and individuals can use layered safeguards to make automated attacks harder to succeed. The supplied article reports security headlines but does not recommend defenses against AI-driven attacks. Useful general measures include multi-factor authentication, prompt software updates, careful access controls, and reliable backups.

Monitoring can flag unusual login rates, repeated failures, or activity that differs from a user’s normal behavior. Rate limits and temporary blocks can slow repeated attempts, while phishing-resistant sign-in methods can reduce the value of stolen passwords. Organizations should also have a clear way to report suspicious messages and a tested plan for responding to incidents.

No single safeguard stops every threat. Controls work best when they are maintained, matched to the organization’s risks, and checked for gaps. The article’s SharePoint headline notes that Microsoft patches failed to fix the on-premises product before a zero-day attack; it does not provide mitigation advice, so specific response steps should come from current vendor and security guidance.

07

Why does automation let a small number of attackers attempt many more attacks than they could carry out manually?

Automation increases an attacker’s reach because software can repeat routine actions without requiring a person to perform each one. The supplied article discusses several security stories, but it does not explain automated cybercrime or quantify this advantage.

For instance, a person can send messages one at a time, while a program can deliver many messages or process many targets according to set rules. Automation can also sort results for later review. That does not remove the need for people to plan campaigns, provide resources, or respond when systems block activity. Nor does it make every attempt effective.

The practical difference is leverage: a small group can attempt more actions with the same amount of human time. Defenders face a related challenge, since they must identify suspicious activity among legitimate events. Monitoring, account protections, and limits on repeated actions can help reduce the advantage. The article reports phishing and a SharePoint zero-day, but does not say either involved this kind of automation.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web