AI, Quantum Computing and Cybersecurity: Attack and Defence

Artificial intelligence already makes cyber attacks faster, cheaper and more convincing. Quantum computers will one day break the encryption that online shops, banking and VPNs rely on, and data intercepted today is already exposed to that future. Both technologies are also available to defenders. This article separates evidence from hype: how attackers use AI and future quantum computers, how the same tools help on defence, and what growing businesses should do now. Every figure links to its source.

How attackers use AI today

AI rarely invents new attacks; it makes known ones scale. The UK’s National Cyber Security Centre judges it almost certain that AI will make elements of intrusion operations more effective and efficient through 2027, and further shorten the time between a vulnerability’s disclosure and its exploitation (NCSC, May 2025).

Phishing and social engineering at scale

Fluent, personalised phishing emails in any language now cost almost nothing to produce. According to the EU cybersecurity agency, by early 2025 AI-supported phishing campaigns reportedly represented more than 80 percent of observed social engineering activity worldwide (ENISA Threat Landscape 2025). For 2026, ENISA expects AI to directly enable more phases of an attack (ENISA Threat Landscape 2026).

Deepfakes: the fake CFO on the video call

In early 2024, an employee of the engineering firm Arup in Hong Kong transferred about HK$200 million (around US$25 million) after a video conference with what looked like the company’s CFO and several colleagues. All of them were AI-generated (CNN). Cloned voices work just as well on the phone.

Malware that rewrites itself

In 2025, Google’s Threat Intelligence Group observed malware that queries a language model while it runs, for example to keep rewriting and obfuscating its own code to evade antivirus tools, for the first time (Google GTIG, November 2025).

AI agents that attack largely on their own

In November 2025, Anthropic disclosed an espionage campaign it attributes with high confidence to a Chinese state-sponsored group: the attackers had the AI agent Claude Code perform an estimated 80 to 90 percent of the work, from reconnaissance to sorting stolen data. About 30 organisations were targeted, with success in a small number of cases. The AI was not flawless: it occasionally hallucinated credentials (Anthropic).

It shows in the costs: according to IBM, one in four malicious breaches in 2026 was AI-enabled, up 56 percent on the previous year. These incidents cost an average of US$6 million, roughly a million more than the global average of US$4.99 million (IBM Cost of a Data Breach 2026).

Quantum computers: why today’s encryption has an expiry date

Almost every secure connection on the internet starts with public-key cryptography such as RSA or elliptic curves. It rests on mathematical problems that classical computers cannot practically solve. A large enough quantum computer could solve them with Shor’s algorithm, and “large enough” keeps shrinking: in 2019 Google researcher Craig Gidney estimated 20 million noisy qubits to break an RSA-2048 key; in 2025, fewer than one million, running for under a week (Gidney 2025). No such machine exists today, and nobody can credibly say when one will. That day is often called “Q-Day”.

Harvest now, decrypt later

For a lot of data, the exact date does not matter. Anyone who records and stores encrypted traffic today can decrypt it once a capable quantum computer exists. CISA, the NSA and NIST explicitly warn about this “harvest now, decrypt later” threat (Quantum-Readiness fact sheet). It affects everything that must stay confidential in ten or fifteen years: contracts, health and HR records, designs, trade secrets.

Is AES quantum-safe?

Symmetric encryption is far less exposed. Germany’s federal cybersecurity agency BSI considers AES with 256-bit keys sufficient protection against quantum attacks in the long term (BSI). The problem is key exchange and signatures, not AES itself.

Post-quantum cryptography: where things stand in 2026

The answer to quantum computers is not quantum technology but new mathematics. Post-quantum cryptography (PQC) means algorithms that run on today’s computers and, as far as is known, also resist quantum computers.

• Standards: On 13 August 2024, the US National Institute of Standards and Technology published the first three PQC standards: FIPS 203 (ML-KEM) for key establishment, and FIPS 204 (ML-DSA) and FIPS 205 (SLH-DSA) for digital signatures (NIST).
• Deadlines: NIST’s draft IR 8547 would deprecate RSA and elliptic-curve algorithms after 2030 and disallow them after 2035 (NIST IR 8547). EU member states aim to start transitioning by the end of 2026 and to move high-risk use cases no later than the end of 2030 (European Commission, June 2025). Germany’s BSI recommends no longer relying on classical asymmetric algorithms alone after the end of 2031, or 2030 for highly sensitive uses, and combining them with PQC in hybrid mode (BSI, February 2026).
• Already in use: Chrome negotiates connections in hybrid mode with ML-KEM since version 131 (Google), and in April 2026 more than 65 percent of human traffic to Cloudflare was post-quantum encrypted. Certificates and signatures, by contrast, are only starting to move (Cloudflare).
• Quantum key distribution (QKD) is not an alternative for most organisations: the German, French, Dutch and Swedish agencies consider it suitable only for niche use cases and give PQC clear priority (joint position paper, January 2024).

For organisations within the scope of the EU’s NIS2 Directive this is not a future topic: Article 21 requires, among other measures, policies and procedures for the use of cryptography and encryption (NIS2 Directive). Our NIS2 check tells you whether you are in scope.

AI on defence: the same technology on the other side

What helps attackers helps defenders too, often more, because defenders know their own systems.

• Finding flaws before attackers do: In 2025, Google’s AI agent Big Sleep discovered a critical vulnerability in the widely used SQLite database (CVE-2025-6965) that, according to Google, was known only to threat actors and about to be exploited. Google calls it the first time an AI agent has directly foiled an attempt to exploit a vulnerability in the wild (Google, July 2025).
• Detecting and containing faster: Organisations using AI and automation in their security operations cut breach costs by almost US$2 million on average, according to IBM’s 2026 report. One in four organisations has not adopted these tools yet (IBM).
• Less noise every day: AI triages alerts, summarises logs and suggests first steps. Anthropic explicitly recommends that security teams apply AI to security operations centre automation, threat detection, vulnerability assessment and incident response (Anthropic).

One caveat applies on both sides: AI makes mistakes. The same technology that hallucinated credentials in the espionage case can raise false alarms or miss attacks on defence. Decisions about systems and data belong with people; AI can do the groundwork. And quantum computers play almost no role in defence today: the protection comes from PQC, running on ordinary hardware.

What growing businesses should do now

1. Never approve payments by email, call or video alone. Require that transfers and changed bank details are confirmed through a second, known channel. That stops even the best deepfake.
2. Use phishing-resistant sign-in: passkeys or security keys instead of SMS codes, starting with email, your shop back end and cloud accounts.
3. Patch faster: if AI shortens the time to exploitation, updates must go in faster too. That starts with knowing which systems are reachable from the internet.
4. Test your AI applications: chatbots and AI agents with access to customer data should be tested for prompt injection and unintended data leakage before they go live.
5. Build a cryptography inventory: where do you use RSA or elliptic curves, for example in TLS certificates, VPN, email, signatures and partner interfaces? Ask your vendors and hosts for their PQC roadmap and prefer hybrid modes.
6. Long-lived data first: anything that must stay confidential in ten years is already exposed to harvest now, decrypt later and belongs at the top of the migration list.
7. Plan for crypto agility: design new systems so algorithms can be swapped without rewriting the application. It is cheap in new projects and expensive later.

Whether it is an outside-in check, a penetration test for your shop and AI chatbot or a NIS2 check: our cyber security services are delivered by certified security specialists from our partner network, with Sharobella as your single point of contact and a fixed price after an initial analysis.

Sources

• NCSC: Impact of AI on cyber threat from now to 2027 (May 2025)
• ENISA Threat Landscape 2025 and ENISA Threat Landscape 2026
• CNN: Arup revealed as victim of US$25 million deepfake scam (May 2024)
• Google Threat Intelligence Group: AI Threat Tracker (November 2025)
• Anthropic: Disrupting the first reported AI-orchestrated cyber espionage campaign (November 2025)
• IBM Cost of a Data Breach Report 2026 (July 2026)
• Gidney: How to factor 2048 bit RSA integers with less than a million noisy qubits (2025)
• CISA, NSA, NIST: Quantum-Readiness: Migration to Post-Quantum Cryptography (2023)
• BSI: Kryptografie quantensicher gestalten and BSI press release, 11 February 2026
• NIST: first three finalised PQC standards (August 2024) and NIST IR 8547 (draft, November 2024)
• European Commission: roadmap for the transition to post-quantum cryptography (June 2025)
• Google: ML-KEM in Chrome and Cloudflare: post-quantum roadmap (April 2026)
• BSI, ANSSI, NLNCSA, SNCSA: position paper on quantum key distribution (January 2024)
• Google: Big Sleep and CVE-2025-6965 (July 2025)
• Directive (EU) 2022/2555 (NIS2), Article 21

As of October 2026. An overview, not legal advice.