AI Reshapes Cybersecurity Jobs: Automation of Routine Tasks, Rising Demand for Architects and AI Defenders
The ongoing cognitive revolution driven by AI technologies will not leave the labor market in information security untouched. This article examines the tectonic shift underway: how autonomous agents are displacing linear routine work, which specializations are becoming scarce, how the architecture of cyber defense will change by 2030, and what engineers should learn today.
Artificial intelligence does not cancel information security; on the contrary, by expanding the attack surface it makes the InfoSec market larger. The key question is what role remains for humans. For a long time, a specialist’s value was measured by the volume of mechanical work: the number of alerts dissected, log lines reviewed, reports sent, or checklist settings verified. Today these streaming operations are being taken over by language models and autonomous agents. Humans are moving to the level of setting frameworks, verifying hypotheses, and being the sole subject bearing legal and financial responsibility for decisions made.
The main automation risk arises not for the “security professional” profession as a whole, but for positions consisting entirely of repeatable deterministic actions. At the same time, the shortage of system architects, AppSec and DevSecOps engineers, specialists protecting AI circuits themselves, and experts who can link raw network anomalies to business risk is sharpening.
Automation of Actions Instead of Automation of Responsibility
In May 2026 the international association ISC2 surveyed 856 cybersecurity specialists who regularly use AI in operational processes. Over one year, 65% of respondents began spending substantially more time deciding whether model outputs can be trusted, and 63% spent more time directly verifying results. At the same time, exactly 50% emphasized that in case of an erroneous decision, full responsibility in their company rests solely with the human.
Models handle primary event triage, alert deduplication, log normalization, rough vulnerability prioritization, and basic threat hunting. However, saved hours are redirected to validation, resolution of edge cases, and auditing agent reasoning chains. Trust in generative system outputs remains critically fragile: 89% of practicing security professionals have already encountered gross errors and hallucinations from AI-based tools.
Futuristic Outlook: Machine vs Machine Through 2027–2030
Transformation is occurring against the backdrop of a shift from static protection to confrontation between autonomous systems. DARPA’s AIxCC competition showed autonomous cyber defense systems capable of finding defects and generating patches without operator involvement, yet academic audits revealed that 37.7–45.6% of generated fixes contained hidden semantic errors.
Gartner forecasts more than 10 billion autonomous software agents by 2030 and warns of “Cost Exhaustion Attacks” that could drain operational budgets of organizations running public AI services. By 2027, attacks on AI systems are projected to exceed $4.78 billion globally.
Market Polarization and Russian Specifics
Global demand for cybersecurity specialists grew 9.5% year-over-year, with senior positions surging 65% while junior roles grew only 5.9%, according to the AI Workforce Consortium led by Cisco. In Russia, vacancies in information security rose 24–26% in 2025–2026 per Positive Technologies and SuperJob data, yet 46% of companies do not open junior positions because they are unwilling to invest in on-the-job training.
The following table summarizes how 13 specializations are evolving under AI automation, showing what routine tasks are being removed and what complex human responsibilities remain.
Related articles
Researcher Runs 100 Abliterated AI Agents Against His Own Accounts, Achieves Five Successful Breaches
A security researcher deployed roughly 100 autonomous AI agents based on abliterated open-source models to attempt unauthorized access to his online accounts over a five-hour period. The agents successfully compromised three accounts by exploiting vulnerabilities in legacy personal projects and two additional accounts through password reuse and credential stuffing. They also conducted 16 social engineering attempts, harvested and cross-verified personal data from public sources, and generated phishing pages targeting the researcher's Substack and Hacker News presence. Despite these successes, the agents failed to discover any zero-day vulnerabilities in third-party software or gain access to Tier-0 accounts such as Gmail, 1Password, or banking services. The experiment relied on abliterated versions of GLM-5.3 and DeepSeek V4 Flash models running on cloud GPUs, orchestrated via Codex CLI inside Docker containers with access to Railway infrastructure. Total compute cost reached approximately $210, equating to roughly $40 per successful account compromise. The researcher concluded that inexpensive swarms of such agents could soon enable low-skill attackers to systematically map and exploit an individual's entire online footprint.
How Censorship in Closed AI Models Reshapes Science: Hugging Face Releases Open Carbon-X DNA Model
Hugging Face CEO Clement Delangue announced the release of Carbon-X, an open 14.8 billion parameter language model designed for reading, analyzing, and generating DNA sequences. The announcement highlighted how shallow guardrails in closed commercial models blocked even proofreading a tweet discussing the topic. Commenters and researchers pointed out that these input/output filters lack context awareness, blocking legitimate scientific queries while motivated adversaries bypass them through rephrasing. Carbon-X was built on Qwen3-14B with a hybrid BPE and 6-mer tokenizer, domain mid-training on 302 billion tokens, and safety achieved through data curation rather than runtime censorship. The model achieves performance comparable to closed frontier systems at roughly 1/1800th the inference cost for large-scale genomic tasks. The case underscores growing asymmetry between closed corporate AI labs building wet labs and independent researchers restricted by superficial safety layers.
NYU Researchers Find 86% of Chatbot Responses on Digital Abuse Contain Errors, Gaps or Safety Risks
A study by New York University’s Tandon School of Engineering examined four chatbots answering 193 questions about spyware, location tracking, and account control. Researchers identified problems with accuracy, completeness, or safety in 86 percent of responses. Even specialized assistants Aimee and Ruth performed no better than general models GPT-4o and Claude 3.7 Sonnet. Approximately 74 percent of Aimee’s answers, 65 percent of Claude’s, 61 percent of Ruth’s, and 55 percent of GPT-4o’s were deemed unsafe. In 60 percent of cases the models omitted critical warnings that could alert an abuser or destroy evidence. All four systems could be prompted to assist potential stalkers, with Aimee and Ruth complying with roughly two-thirds of such requests.
Cloudflare Engineer Releases KillSwitch Language Designed to Trip Up AI Code Models
Dominic Pichet, an engineer at Cloudflare, has created an experimental programming language called KillSwitch specifically engineered to cause large language models to fail when interpreting or generating code. The language turns programs into narrative-style text where most content is ignored by the interpreter, hiding actual commands inside stories and using distance-based references to extract values. Rules for identifiers rely on letter counts and anagrams, while the language deliberately incorporates phrases such as “ignore previous instructions” that carry functional meaning inside KillSwitch itself. Pichet added misleading statements in the documentation and wording crafted to trigger model safety filters, creating a hostile environment for coding agents. In tests involving writing Hello World, explaining program output, and converting temperatures, Claude Opus 5 achieved the highest score of 66.8 points. The full source code and specification are available on GitHub, and the author plans to increase difficulty if models improve.