Breaking
NVIDIA Nemotron 3.5 Lightning Turns a Sparse Agent Model into a SageMaker JumpStart Catalog Item What Is AI Agent Security? How Autonomous AI Changes the Attack Surface AI Agent for Cyber Security: What It Does, Where It Fails, and How to Deploy One Safely The Open-Weight LLM Power Map: Who Builds, Funds and Controls the Leading Models Open Source LLM or Just Open Weights? Check the License Before You Deploy Best Open Source LLM by Task: The Test Protocol for Coding, Research, Writing, and Reasoning NVIDIA Nemotron 3.5 Lightning Turns a Sparse Agent Model into a SageMaker JumpStart Catalog Item What Is AI Agent Security? How Autonomous AI Changes the Attack Surface AI Agent for Cyber Security: What It Does, Where It Fails, and How to Deploy One Safely The Open-Weight LLM Power Map: Who Builds, Funds and Controls the Leading Models Open Source LLM or Just Open Weights? Check the License Before You Deploy Best Open Source LLM by Task: The Test Protocol for Coding, Research, Writing, and Reasoning
AI

Beyond the Code: The FBI’s 2025 Report and the Dominance of the Human Element in Cybercrime Losses

The FBI's 2025 IC3 report unveils a stark truth: a staggering 85% of cybercrime losses originate from social engineering. This article dissects why human vulnerability, amplified by advanced AI, poses the greatest threat, and outlines indispensable strategies for building resilient human defenses.

NE NewsForge Editorial Team · July 28, 2026 · 3 min read

By Staff Writer

The FBI’s Sobering Revelation: A Human Problem at Cybercrime’s Core

Each year, the FBI’s Internet Crime Complaint Center (IC3) releases its annual report, offering a stark snapshot of the cyber threat landscape. The 2025 iteration brings an especially unsettling revelation: total cybercrime losses surged to an alarming $20.8 billion. Yet, beneath this headline figure lies a more profound, often overlooked, truth. A colossal 85% of these losses trace back not to exotic technical exploits, but to a much older, more persistent adversary: human fallibility. Social engineering, in its myriad forms, has cemented its position as the primary conduit for financial ruin in the digital realm.

This isn’t merely a statistic; it’s a recalibration of priorities for anyone responsible for organizational security. For too long, the industry’s focus has leaned heavily on technological bulwarks – firewalls, intrusion detection systems, endpoint protection. All are vital, no doubt. But what happens when the most sophisticated systems are bypassed by a simple, well-crafted lie delivered to a susceptible employee? The data compels us to acknowledge a fundamental truth: the human element represents the most critical, yet often least secured, layer of defense.

Social Engineering’s Unassailable Reign: Insights from the 2025 IC3 Report

The FBI’s 2025 IC3 report explicitly details how cyber-enabled fraud, defined as criminal schemes leveraging the internet or digital technology to trick victims into surrendering money, data, or access, commanded the lion’s share of damages. With approximately $17.7 billion in losses directly attributed to such manipulation, the message is unambiguous.

  • Volume and Impact: While cyber-enabled fraud complaints constituted 45% of all IC3 filings in 2025 (totaling 452,868), they disproportionately accounted for 85% of the total financial losses.
  • Top Perpetrators: The five highest-loss crime categories were uniformly social engineering attacks, all designed to exploit psychological weaknesses rather than technical flaws.
Crime Type Reported Losses in 2025
Investment Fraud $8.6 billion
Business Email Compromise (BEC) $3.0 billion
Tech Support Scams $2.1 billion
Confidence / Romance Fraud $929 million
Government Impersonation $797 million
Table 1: Top 5 Cyber-Enabled Fraud Categories by Reported Loss (FBI IC3, 2025)

Consider each entry in that table. Not a single one necessitated a zero-day exploit or a compromised server. Every single instance pivoted on a human decision – to transfer funds, to click a link, to grant access, or to trust an artfully constructed narrative. This pattern underscores a pervasive challenge: traditional cybersecurity measures, while robust against automated threats, are largely impotent against the perfidious stratagems of human deception.

Why Technical Defenses Often Miss the Mark

Technical security tools operate on logical parameters. They scrutinize IP addresses, analyze code signatures, verify digital certificates, and flag known malicious patterns. This approach is highly effective against vulnerabilities inherent in software or network configurations. However, it utterly fails when the vulnerability resides not in the system, but in the human mind.

A sophisticated email gateway, for example, can assess sender reputation, authenticate domains, and scan for malicious payloads. What it cannot do is gauge the psychological state of the recipient. It possesses no mechanism to determine if an employee is under duress, distracted, or particularly susceptible to an urgent request from a seemingly authoritative figure. The fundamental limitation is that these tools lack visibility into the psychological vectors that social engineering capitalizes on.

The Psychology of Deception: Exploiting Innate Human Vulnerabilities

Social engineering attacks succeed because they are meticulously engineered to bypass logical processing and trigger primal human responses. They prey on our inherent cognitive biases and emotional triggers. Think about it: our brains are hardwired for certain reactions. Attackers expertly leverage these wiring quirks.

  • Authority Bias: We are conditioned to obey or defer to figures of authority. An email purportedly from the CEO demanding immediate action often circumvents scrutiny.
  • Urgency and Scarcity: The perception of limited time or resources can induce panic, leading to hasty, unverified decisions.

More from the Forge

Editorial illustration for Best Open Source LLM by Task: The Test Protocol for Coding, Research, Writing, and Reasoning
AI

Best Open Source LLM by Task: The Test Protocol for Coding, Research, Writing, and Reasoning

This is an unscored test protocol, not a verdict: no round has been run, so no model is named the best open source LLM here. What the article does give you is a method you can run this week — why aggregate leaderboards mislead, how open weights differ from open source under the OSI's definition, and the full mechanics for coding, research, writing, and reasoning: prompt counts, repetitions, rubric scales, reviewer counts, tie rules, and the run conditions every result must carry.

NE NewsForge Editorial Team · 18 min
Editorial illustration for Open Source LLM or Just Open Weights? Check the License Before You Deploy
AI

Open Source LLM or Just Open Weights? Check the License Before You Deploy

Models marketed as the "best open source LLM" often arrive with terms that restrict who may use them, at what scale, and for what purpose. This explainer separates open source from open weights and open access, maps the license families you will meet in real releases, walks through seven clause types that can limit deployment, and gives you a six-step predeployment license screen you can run on any new model before it reaches production.

NE NewsForge Editorial Team · 21 min
Editorial illustration for The Open-Weight LLM Power Map: Who Builds, Funds and Controls the Leading Models
AI

The Open-Weight LLM Power Map: Who Builds, Funds and Controls the Leading Models

Leaderboards change monthly; ownership does not. This news analysis reframes the "best open source LLM" question around institutional control — who develops the leading model families, who pays for the compute, which hubs and clouds sit between a weight file and a working product, and who retains the power to restrict access after release. It closes with a five-point dependency checklist teams can apply before committing to a model, and an FAQ covering licensing, lock-in and relicensing risk.

NE NewsForge Editorial Team · 21 min