Economy

Add as a preferred source on Google

Andreessen Horowitz sees AI reviving an earlier era of cyberattacks

Artificial intelligence is making software attacks faster and cheaper, reviving the emergency conditions of earlier cybersecurity eras, venture capital firm Andreessen Horowitz said in a Nov. 16 newsletter.

Updated 3 min read
Andreessen Horowitz. (Photo: Vice-Presidência da República (VPR) / CC BY 2.0, cropped)
Andreessen Horowitz. (Photo: Vice-Presidência da República (VPR) / CC BY 2.0, cropped)

Artificial intelligence is making software attacks faster and cheaper, reviving the emergency conditions that defined earlier decades of cybersecurity, venture capital firm Andreessen Horowitz said in a newsletter published Nov. 16.

The firm, which invests in startups, said software repairs have not kept pace with attackers’ growing capabilities. Its newsletter, “The Dream of the 1990s Is Alive Again in Cybersecurity,” argued that the gap could revive attacks that spread automatically between vulnerable computers.

The analysis drew on conversations with security teams over the past several months and comparisons with attacks in the 1990s and 2000s. It provided no numerical estimates of the current increase in vulnerabilities or attack speeds.

Andreessen Horowitz said security emergencies that once interrupted teams every several years had occupied much of 2026 for teams it spoke with. The firm linked the change to Opus 4.5 and 4.6, saying those models reduced the expertise and time needed to find and exploit software flaws.

The authors said the resulting risks extend to both zero-days, previously unknown software flaws, and older vulnerabilities that organizations have left unfixed. They compared the situation with 2002, when serious flaws affected Apache, OpenSSH, OpenSSL and BIND, software supporting websites and other basic internet functions.

The newsletter predicted a return of internet worms, malicious programs that copy themselves from one vulnerable computer to another. Attackers could use them to demand ransom or steal access credentials for artificial intelligence services, giving the attacks commercial purposes beyond simply spreading.

The authors also expected businesses to install security updates much faster as the risks of waiting increase. But the volume of corporate software, including programs no longer actively maintained, makes that difficult. Companies have also delayed automatic updates because changes can disrupt operations.

Andreessen Horowitz expected new businesses to offer more secure versions of essential infrastructure software. It acknowledged that earlier companies pursuing that approach struggled commercially and left open whether the current changes would create a larger market.

The authors called for continuous security testing that imitates attackers, including tests inside corporate networks. Many teams they spoke with were shifting toward continuous testing and treating older, annual testing services primarily as a way to meet compliance requirements.

The newsletter argued that businesses need overlapping attack detection and faster responses because AI-assisted attackers could defeat security controls even when those controls work correctly. It also called for changes to software that protects employee devices, manages access and checks outside software packages.

The firm said businesses now need controls governing software agents’ actions alongside those governing people. It cited Neo, Keycard and Socket as companies addressing parts of that shift. The supplied text ended mid-sentence during a discussion of the time between discovering a vulnerability and developing an attack.

Related articles

View all