The war between businesses and hackers enters a perilous new phase

The cybersecurity landscape is entering a dangerous new phase as artificial intelligence accelerates the capabilities of both attackers and defenders. Anthropic and OpenAI recently unveiled AI models—Mythos Preview and 5.5-Cyber—that are so effective at bypassing defenses that they were only shared with trusted firms. Yet even this controlled access offers no guarantee of safety, as hackers already exploit earlier AI tools, and the asymmetry of the fight favors attackers. A single successful breach by an AI-powered hacker can succeed where defenders must remain flawless. Cybersecurity firms and businesses are scrambling to adapt, forming alliances like those between Palo Alto Networks (PAN) and CrowdStrike to mitigate risks. Initial reactions to the new models included panic, but they also served as a wake-up call, pushing cyber-readiness discussions from IT departments to boardrooms. Experts warn that pessimism currently outweighs optimism, as criminals inevitably catch up with restricted AI advancements. The British government’s AI Security Institute suggests OpenAI’s 5.5 series may even surpass Mythos in capability, raising concerns about rapid advancements from open-source or Chinese developers. Before the latest models emerged, AI already amplified hacking threats. CrowdStrike reported an 89% increase in AI-enhanced attacks in 2025, while PAN found that the fastest attackers now steal data in under an hour—down from nearly five hours in 2024. Many firms take days to detect breaches, leaving them vulnerable. Autonomous AI agents compound the risk, enabling hackers to execute multi-tasking attacks while businesses adopt AI for coding, customer service, and other functions, expanding attack surfaces. PAN’s recent acquisition of Portkey, a firm specializing in AI agent security, reflects the urgency of the challenge. Jeremy D’Hoinne of Gartner notes that companies are building more software than ever, increasing exposure to vulnerabilities. The number of reported “common vulnerabilities and exposures” (CVEs) is surging, with some predicting annual detections could rise tenfold to 480,000 as AI identifies flaws faster than patches can be developed. This explosion of vulnerabilities strains already overstretched security teams, deepening the crisis. The proliferation of AI-driven cyber threats is reshaping the industry, demanding immediate action from both security firms and corporate leadership. While alliances and acquisitions signal a coordinated response, the scale of the challenge—combining autonomous hacking tools, unchecked vulnerability growth, and the race between attackers and defenders—poses a perilous new frontier for global cybersecurity.