PwC survey: firms unready for AI and quantum threats

PwC survey: firms unready for AI and quantum threats

Companies around the world are adopting AI faster than they are learning to protect it. Few are preparing for the day quantum computers can break today's encryption. These are the main findings of PwC's 2027 Global Digital Trust Insights report.

The consultancy surveyed nearly 4,000 business and technology leaders in more than 70 countries. The goal was to see how organizations are responding to a fast-changing digital environment. AI is at the center of almost every finding.

Attacks on AI systems top the worry list

Respondents expect AI-assisted attacks to grow in volume and speed. The more notable result is that they named attacks on their own AI systems as the threat they are least prepared to handle.

The scenarios they feel least able to defend against are:

  • compromise by autonomous botnets (53%)
  • adversarial attacks (52%)
  • data poisoning (52%)

Prompt injection belongs in this picture as well. In late 2025, OpenAI said that prompt engineering, like social engineering, is unlikely ever to be fully solved. When attackers abuse it, prompt injection can leak data or run unauthorized commands, and there may be no reliable way to prevent it. It also ranks first in this year's OWASP top ten list of threats to LLM applications.

Defenders want AI, but with a human in charge

Leaders also see AI as the top defensive tool for threat detection and alerting. That trust has clear limits. Only 22% would accept "fully autonomous execution by AI agents for cyber defense without human approval."

Asked why they hold back, 55% pointed to problems with reliability and maturity. Another 44% cited a lack of staff skilled in AI oversight and governance.

SecurityWeek's Kevin Townsend doubts this caution can last. AI-assisted attacks are gaining speed, scale and sophistication faster than humans can respond. Only autonomous defense can match that pace, and an agent that waits for human approval is not truly autonomous. Organizations may eventually have to choose between fully autonomous agents and repeated compromise. Full autonomy brings its own cost, because it widens the attack surface.

There is also no agreement on who is responsible for AI:

  • 29% place accountability with the CIO, CTO or technical function
  • 26% assign it to a dedicated, named AI leader or function
  • 17% give it to the CISO or cyber function
  • 11% share it across several functions

Budgets are moving in the same direction. Of the security and finance leaders surveyed, 84% expect spending to rise, and 58% rank AI as their top cyber budget priority.

Encrypted data on borrowed time

The report also covers data protection. AI makes it easier for attackers to steal data. Encryption has long been the standard fallback, because stolen data is of little use if it cannot be read.

Quantum computing threatens that safeguard. Quantum computers running Shor's algorithm are expected to break classic encryption, a moment often called "Q-day." Attackers are already taking encrypted data and storing it until they can decrypt it, a tactic known as "harvest now, decrypt later."

Quantum-resistant cryptography is available and can replace current encryption. PwC says the time to prepare is now, yet only 21% of respondents are implementing quantum-resistant security measures.

"Technology is moving incredibly fast, but the fundamentals of cybersecurity haven't changed," said Morgan Adamski, PwC's cyber, data and technology risk leader. "You can invest heavily in AI and the latest security tools, but if you don't have secure data, operational continuity, clear accountability and strong cyber hygiene underneath them, you're building on a weak foundation."

A snapshot, not ground truth

Townsend adds some caveats about surveys like this one. They capture opinions at one point in time, and AI is changing fast enough that those views can be out of date by publication. The questions and answers are subjective. He also warns that the results could reduce urgency: a company that knows it is behind may relax after learning that many peers are in the same position.

Our Take

For readers, the most concrete number in the report is probably the 21% figure on quantum-resistant measures. "Harvest now, decrypt later" means data stolen today can be exposed years from now. Infrastructure providers are starting to act, for example Cloudflare's plans for post-quantum certificates in 2027. This suggests the tools exist and that adoption depends on organizational will more than on technology.

The debate over AI autonomy is likely to get sharper. Recent incidents such as the DIVD breach by an autonomous AI agent and the faster exploitation of vulnerabilities support Townsend's argument that human-paced defense may fall behind. It is worth watching whether next year's survey shows more trust in autonomous agents, and whether accountability for AI moves closer to security teams. Without that clarity, larger budgets may not lead to stronger defenses.