The IA eliminates the technical barrier: automated attacks that anyone can execute and the defense that must be reinvented

In 2025 we saw scenes that previously appeared to be taken from novels or from old hackers' records: adolescents without technical training capable of removing millions of records, attacks by a single actor who would have previously required organized equipment, and malicious packages invading open-source ecosystems on an unpublished scale. The key lesson is that artificial intelligence has not invented new motives for crime, but it has removed the technical barrier that previously separated the intention from action.

The indicators speak with alarming clarity: the count of malicious packages detected in public repositories jumped from tens of thousands to hundreds of thousands in a few years, and metric as the time from the spread of a vulnerability to its exploitation has been reduced from almost two years to weeks or even days, according to the sector's reports. These trends are not anecdotal: response organizations and public intelligence providers have documented that exploits appear almost at the same time that vulnerabilities are disclosed, which completely disrupts the traditional "pitcher window" assumptions. To deepen the magnitude of the problem, industry resources such as Sonatype's analysis of the software supply chain and incident response reports from threat intelligence providers can be consulted ( https: / / www.sonatype.com /, https: / / www.mandiant.com /).

Why did this happen? Large language models and agentic platforms automate technical steps that previously required expertise: generating functional code, creating exploits, avoiding detection and even automation of extortion or social engineering campaigns. The result is that actors with banal motivations - from virtual goods purchases to small economic rewards - can orchestrate attacks with professional effectiveness. The threat ceased to be just a question of "who has talent" and became "who has access to powerful tools and time to use them."

In addition to capacity, the form of the threat changed. Malicious packages now include documentation, unit tests and structures that mimic legitimate projects, undermining the classic tools of static detection and signature. Organizations feel the pressure: code stops, secret commitments and economic losses from malware proliferation in public dependencies. This reality shows that the specific defenses and the race to speed up patches will not be enough on their own.

Against this background, the defensive strategy must evolve and prioritize structural measures that reduce the exploitable area. It is not just about parking faster, but about neutralize entire categories of attack where possible:: implement policies that require verifiable supply of packages, produce and consume reconstructed artifacts from assigned sources, incorporate chain of custody signatures and verifications into CI / CD pipelines, and support distribution mechanisms that prevent the implantation of packages and the poisoning of dependencies. Initiatives such as the signing of devices and trust repositories, as well as specific tools to protect the supply chain, are practical supplements to these policies ( https: / / www.chainguard.dev /).

At the same time, organizations should strengthen operational controls that limit impact when an intrusion occurs: automatic and segmented rotation of secrets, minimum design privileges, separation between development and production environments, cataloguing and governance of dependencies with verifiable SBOMs, and patching automation prioritizing risk and real exposure. Detection and response also require investment: build and run telemetry, supply chain-related alert enrichment and playbooks that look at IA-accelerated scenarios. The authorities and public response centres provide practical guides and frameworks for prioritizing these actions ( https: / / www.cisa.gov /).

Not everything is technology: governance, responsibility and culture matter. Realistic training programmes for developers, clear policies on the use of coding assistants in corporate repositories, comprehensive unit governance reviews and contractual agreements with software providers can mitigate systemic risks. And, where appropriate, legal and enforcement teams must work with operations to reduce friction that prevents rapid and complete mediation.

Looking forward, the combination of more powerful models and a growing rate of software production anticipates that the pressure will not decrease. Effective defense requires moving the approach from "being faster than the attacker" to "making certain types of attacks impossible" by monitoring integrity, source verification and designs that reduce the dependence on unverified components. Adopting these practices will not stop all threats, but it will make cheap and automated attacks much less cost-effective and much more detectable, which is just what security needs to recover ground in this new era.