Hundreds of Millions of Records, One Threat Actor and an API Nobody Was Watching
Between late December 2025 and mid-February 2026, a single threat actor breached at least nine Mexican government organizations across federal, state, and municipal levels, including SAT, Mexico City’s civil registry, the national electoral institute, and four state governments. Gambit Security reported that the SAT breach exposed roughly 195 million taxpayer records and domain-wide credentials across 305 servers, while Mexico City’s civil registry lost about 220 million records. The attacker also built a live REST API into SAT’s production tax systems to generate forged tax certificates using real government data.
What made the campaign notable was the role of AI: an AI coding assistant generated roughly 75% of remote command execution, taking the attacker from initial access to working exploit code on a live government server in about 40 minutes. A separate AI model processed reconnaissance from the 305 compromised servers into thousands of structured intelligence reports. Anthropic confirmed it investigated the misuse of Claude, banned the accounts involved, and added mitigations.
Two months later, France’s national identity document portal, operated by France Titres (ANTS), confirmed a breach detected on April 15, 2026, affecting roughly 11.7 million accounts and exposing names, dates of birth, addresses, and contact details. A threat actor claiming 18–19 million records offered them for sale, but that figure was not confirmed by ANTS. The technical cause remains unconfirmed: researchers have pointed to an IDOR/API authorization flaw, while compromised API credentials or a third-party integration remain alternative possibilities.
What connects both cases is not a stolen password or phishing email, but API-level exposure of sensitive government data, and in both, the weakness was less an exotic exploit than a system that lacked sufficient visibility and protection. Hence why we are diving deep into the issue of API Visibility in Government Infrastructure.
Why Governments Have Become the Softest Hard Target
Government agencies hold something attackers can’t get anywhere else at the same scale: permanent identity data. A stolen credit card gets cancelled. A stolen Social Security number, tax history or biometric record doesn’t expire; it can be used for identity theft and espionage for decades.
That single fact reshapes the threat model. Where cybercriminals target ecommerce APIs for a quick financial payout, public sector APIs draw a different class of adversary: persistent, well-resourced actors and nation-states after long-term access, not a one-time transaction.
Three structural pressures compound the exposure:
- Legacy debt. Agencies rarely get to build from scratch; developers are routinely tasked with wrapping modern APIs around decades-old back-end systems, creating integration points that were never designed with today's threat landscape in mind.
- Zero tolerance for downtime. Where an e-commerce outage frustrates shoppers, a government API failure on emergency dispatch or benefits disbursement disrupts entire communities and cannot simply be taken offline mid-incident to patch.
- A widening blast radius. In India, CERT-In tracked 204,844 cybersecurity-related incidents involving government organisations in 2023, up from 192,439 in 2022. Across all sectors, CERT-In handled more than 1.59 million cybersecurity incidents in 2023 alone, illustrating the scale of India's cyber-threat landscape. However, these figures represent reported or handled cybersecurity incidents and should not all be described as successful attacks.
The pattern is also evident in the United States. Federal agencies reported 32,211 information-security incidents in FY2023, a 9.9% increase from the previous year. Separately, a Comparitech analysis found that ransomware incidents targeting government entities worldwide rose 65% year-over-year in the first half of 2025, reaching 208 incidents. FBI data have also identified government facilities as a major ransomware target: in the FBI’s 2022 critical-infrastructure data, government facilities ranked third by reported ransomware incidents, behind healthcare and critical manufacturing.
The Vulnerability Class Static Tools Were Never Built to Catch
Traditional firewalls look for known attack signatures, malformed requests, blacklisted payloads, and obvious injection strings. Modern API abuse rarely looks like that. It looks like legitimate user behavior manipulated at scale, which is exactly why legacy security tools miss it.
Three vulnerability classes dominate public-sector incidents today:
- Broken Object Level Authorization (BOLA). One of the most persistent API vulnerabilities: an endpoint accepts a user-supplied ID without checking whether the requester is authorized to access that record. Manipulating IDs can let attackers systematically extract citizen data, a flaw researchers believe may have caused the ANTS breach, though French authorities have not confirmed the root cause.
- Business logic abuse. Attackers exploit how an API is designed to work rather than a coding flaw—for example, quietly scraping a voter registry through legitimate-looking requests. To a traditional WAF, each request may appear normal until the data has already been exfiltrated.
- Shadow and zombie APIs. Forgotten test environments, undocumented endpoints, and outdated API versions can create unmonitored paths into agency networks. The risk grows as agencies deploy AI agents and Model Context Protocol (MCP) servers, generating new, often untracked traffic patterns that security teams may not have inventoried.
Government-specific research backs this up directly. Public APIs used for interagency data feeds and citizen-facing services are flagged as a top exposure category, where broken authorization or excessive data returned in a single response can hand an attacker a full database dump instead of the single record a legitimate query would return.
Why 2026 Compounds the Problem: AI on Both Sides of the Fight
The Mexico campaign wasn’t an isolated curiosity; it previews how AI can accelerate conventional attacks. IBM’s 2026 X-Force Threat Intelligence Index found exploitation of public-facing applications rose 44% year over year in 2025, becoming the leading initial-access vector. X-Force also tracked nearly 40,000 vulnerabilities, with 56% exploitable without authentication—a significant risk for internet-facing citizen services.
AI is accelerating the workflow further, streamlining reconnaissance and vulnerability discovery and lowering the barrier to sophisticated attacks. The Mexico campaign illustrates this shift: according to Gambit Security, one operator used Anthropic’s Claude and OpenAI’s GPT for reconnaissance, exploitation, scripting, and analysis. AI wasn’t a side utility; it functioned as operational tooling throughout the attack.
The Regulatory Signal: Continuous Visibility Is No Longer Optional
Regulators on multiple continents are converging on the same demand: agencies must know what APIs they’re running in real time, not on an annual audit cycle.
In the Verizon DBIR 2026 report, vulnerability exploitation became the leading breach entry point at 31% breaches involving third parties rose from 30% to 48%, a 60% increase. This reinforces the importance of vulnerability management, attack-surface visibility, and third-party/API risk management.
Hence Continuous automated API discovery is becoming the baseline expectation for good API Governance and not just a best practice anymore.
What Government-Grade API Visibility Actually Requires
Public sector environments have constraints private companies don’t: services can’t go dark for maintenance, budgets are fixed years in advance, and IT teams are frequently stretched thin across sprawling, decentralized systems. A workable API security program for the government has to account for all of it.
- Continuous automated discovery, not annual inventories. Manual API inventories are outdated the moment they're written; automated tools that continuously scan traffic to map every active endpoint are what actually expose shadow APIs before attackers find them.
- Zero Trust is enforced at the API layer, not just the network edge. Securing north-south perimeter traffic is a necessary foundation, but a request originating inside a government network is not inherently safe; every API call needs to be authenticated, authorized, and inspected on its own merits, with sender-constrained tokens replacing easily intercepted bearer tokens.
- Behavioral analytics tuned to catch valid-looking abuse. Most API incidents in 2026 happen through valid credentials and expected workflows. Nothing crashes, nothing looks broken, data simply leaks or unauthorized actions occur quietly, which means detection has to baseline normal usage and flag deviation not just block known-bad signatures.
- Identity controls that extend to AI agents, not just humans. As agencies deploy autonomous AI agents to interact with citizens or other systems, identity has to extend to those digital entities too, through defined, verifiable cryptographic permissions comparable to what a human user gets.
- Precise context-aware authorization. Authentication proves who's calling; authorization decides what they can actually do, and most API failures happen at that second step, which is why access decisions increasingly weigh ownership, request context, and data sensitivity rather than a simple role check.
How Prophaze AI-Native WAAP Closes the Gap
A modern Web Application and API Protection (WAAP) platform combines the functions agencies have historically bought separately from WAF, API security, bot mitigation, and runtime threat detection into one layer built for continuous automated operation rather than periodic review.
For public-sector environments, that means:
- Unknown APIs become visible. Runtime API discovery continuously identifies shadow, zombie, and legacy-connected endpoints, helping agencies close the gap between what is actually exposed and what is documented.
- Attacks that look like legitimate traffic get detected. Behavioral threat detection identifies abnormal usage, business-logic abuse, and BOLA-style manipulation that traditional signature-based controls can miss.
- Legacy systems can be protected without taking them offline. Virtual patching protects vulnerable legacy and VPN-accessed applications when agencies cannot wait for a full application upgrade or emergency maintenance window.
- Citizen services stay available during attacks. Layer 7 DDoS protection absorbs application-layer floods and automated traffic so benefits portals, court systems, and other public services can remain accessible.
- Security doesn't become another performance problem. Sub-millisecond decision-making enables real-time inspection and enforcement without adding significant latency to citizen-facing applications.
- Security can fit existing government infrastructure. Deployment across on-premise, cloud, multi-cloud, and Kubernetes environments allows agencies to protect services without forcing a wholesale infrastructure change.
Together, these capabilities give agencies something traditional perimeter security and periodic assessments cannot: continuous visibility, detection, and automated protection across the applications and APIs citizens depend on. Prophaze combines AI-driven behavioral analysis, runtime API discovery, automated protection, and flexible deployment into a single WAAP layer built for the operational and infrastructure constraints of the government.
PROOF IN PRODUCTION: KERALA STATE IT MISSION
When Kerala’s citizen-facing digital services came under sustained Layer-7 attack in 2025, Prophaze deployed its WAAP within the Kerala State Data Centre, using AI-powered bot mitigation, behavioral analysis, and adaptive traffic control. The result:
350M+ malicious requests blocked | 1.5M peak requests handled | 0 sec downtime | 100% service continuity
The protection handled a high-volume attack within a 15-minute attack window, while a critical police application was stabilized in under 10 days. Citizen services remained accessible throughout.
- The Question Every Agency Should Be Asking
Every API an agency adds for benefits, permits, interagency data, or citizen identity expands its attack surface. The Mexico and France incidents show how systems that aren’t sufficiently visible, monitored, or tested can be exploited at scale.
The gap between what’s running and what’s known is where the next breach begins. Closing it starts with visibility.
Frequently Asked Questions (FAQ)
1. What does API visibility mean for government agencies?
It means having a continuously updated, accurate inventory of every API an agency runs, including shadow APIs nobody documented and zombie APIs that were supposed to be retired, along with real-time insight into how each one is actually being used so abnormal or unauthorized activity can be caught before it becomes a breach.
2. Why are government APIs targeted more than typical corporate APIs?
Government databases hold permanent identity data, Social Security numbers, biometric records, tax and health histories that don’t expire the way financial data does, which attracts persistent, well-resourced attackers and nation-states seeking long-term access rather than a quick payout.
3. What is BOLA and why does it matter for citizen portals?
Broken Object Level Authorization occurs when an API checks that a user is logged in but doesn’t verify they’re authorized to view the specific record they’re requesting.
4. How is AI changing attacks on government infrastructure?
Attackers are using AI coding assistants and language models to accelerate reconnaissance, generate exploit code, and process stolen data at a scale and speed that used to require large teams, as seen in the 2025–2026 campaign against nine Mexican government agencies where AI tools drove the majority of the attacker’s technical activity.
5. What should agencies prioritize first to improve API security?
Continuous automated API discovery is the foundational step: agencies can’t secure or test endpoints they don’t know exist. From there, Zero Trust enforcement at the API layer, behavioral anomaly detection, and virtual patching for legacy systems close the highest-impact gaps fastest.
