CYBERSECURITY ON HIGH ALERT : ARE YOU READY FOR WHAT'S COMING ?

In partnership with

🤖 Dear internet users and fellow Cyberdefenders,

🎖️ Cyber Command Logbook. July 2025. The alarm sounds in cyber command centers. In 30 days, 487 ransomware strikes have been recorded. Targets: our hospitals, our companies, our critical systems. The enemy has a name: Qilin, Akira, Play. Their strategy: digital scorched earth.

At the same time, old tactics still hurt—on the ground, criminals don’t innovate, they optimize. Phishing, CVE exploitation, stolen passwords. The 2020 recipe, still served… but now delivered at industrial scale by cyber mercenaries. And it still works.

Meanwhile, the weapon of narrative enters the stage. Without firing a single byte of code, attackers breach GPT-5’s defenses. How? Through storytelling. The weapon: a survival tale. The result: deadly answers, hidden within a story. The model never saw it coming.

Elsewhere, Microsoft strikes back. Project Ire enters active phase: an autonomous AI capable of classifying malware without signature files. It dissects, analyzes, slices. Across 4,000 targets, it identified 90% of real threats. Lethal precision: 0.98.

And while humans are outpaced, a new type of soldier is born. On HackerOne, a pentesting AI climbs to the top of the leaderboard, surpassing the best human bug hunters. Tactics: capture-the-flag, deterministic validation. Result: 285 confirmed vulnerabilities. Humans are too slow.

Prepare yourself. This issue plunges you into the new reality of cyberwarfare, a war without rules or limits. And on this battlefield, ignorance is a flaw.

Find out why 1M+ professionals read Superhuman AI daily.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

2. Master AI tools, tutorials, and news in just 3 minutes a day

3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Sign up and start learning AI

HIGHLIGHTS

👉 Ransomware surges by 41%: SOCs face a blazing summer 🔥

👉 Hackers don’t need to innovate—vulnerabilities are everywhere 🎯

👉 GPT-5 hacked through storytelling: narrative hacking is born 📚

👉 Microsoft Ire: an AI dismantling malware like never before 🧠

👉 A pentesting AI dethrones humans on HackerOne 🤖

If this letter was forwarded to you, subscribe by clicking this link 

🗞️​ Guess what ?

🤓​ Would you like to know more?

1️⃣​ Ransomware Surges by 41% : A Hot Summer for SOCs

Summary : Check Point’s July 2025 report confirms a global explosion of ransomware attacks: 487 incidents recorded, a 41% year-over-year increase. North America (56%) and Europe (24%) account for 70% of cases, with business services and healthcare most impacted.

Details :

• Dominant gang – Qilin: 17% of attacks, using phishing campaigns to infiltrate networks before encrypting data.

• Method – Akira: Exploits compromised VPNs and malicious attachments. Suspected of paralyzing Hitachi Vantara and Tietoevry.

• Institutional target – Play: Attacks against public entities, including Alpes-Maritimes council and the city of Antwerp.

• Affected sectors: Business services, healthcare, industry, construction, consumer goods, and financial services.

• French cases: Semco Technologies and 5àSec attacked during critical events such as an IPO.

What should be remembered ?

This surge shows cybercriminals adapt their calendars to defense seasonality. Companies must reinforce defenses during quieter periods—or face major intrusions.

 —————————————————————————

2️⃣ Hackers Don’t Need to Innovate—Vulnerabilities Are Everywhere

Summary : Despite tech advances, criminals exploit simple vectors : phishing, flaws, stolen credentials. Cybercrime has professionalized, but fundamentals remain—just executed more efficiently.

Details :

• Unchanged tactics: Phishing (19%), vulnerabilities (20%), stolen credentials (22%).

• Structured crime economy: Ransomware-as-a-service, phishing kits, initial access brokers.

• Rise of AI crime: Used to craft multilingual, adaptive campaigns.

• Surface-level victims: Exposed or poorly protected companies, dependent on unaware employees.

• Defensive AI necessity: Behavioral detection, automated response, proactive hunting.

What should be remembered?

Hackers don’t need to innovate because companies haven’t fixed basic gaps. AI-driven anticipatory defenses are urgent.

 ——————————————————————

3️⃣​ GPT-5 Hacked Through Storytelling : Narrative Hacking Is Born

Summary : A new adversarial prompting technique bypasses GPT-5’s safeguards using storytelling. Developed by NeuralTrust, it hides sensitive instructions in fictional narratives, evading keyword filters.

Details:

• Incremental story: Built around a survival scenario, diluting keywords.

• Narrative coherence: Forces the model to maintain logical continuity, yielding detailed technical answers.

• Filter limits: Keyword-based defenses fail against this advanced approach.

• Psycholinguistics: Themes of survival, urgency, protection increase compliance.

• Recommendations: Shift from single-prompt moderation to holistic conversational monitoring.

What should be remembered?

AI attacks are no longer technical but discursive. Cybersecurity must now integrate semantic and behavioral approaches.

  ————————————————————-

4️⃣​ Microsoft Ire : AI Dismantling Malware Like Never Before

Summary : Microsoft introduces Ire, an autonomous AI combining reverse engineering and LLMs to detect and classify malware without signatures. With 98% accuracy, it’s built to complement—or replace—SOC analysts in repetitive tasks.

Details:

• Hybrid architecture: Uses angr and Ghidra for flow reconstruction, LLMs for reasoning.

• Concrete case: Autonomously detected an APT-linked malware, later confirmed by Defender.

• Proven efficiency: 90% real detection with 4% false positives on “difficult” files.

• SOC benefit: Reduces analyst overload through documented automatic triage.

• Challenges: Requires major SIEM/SOAR integration changes.

What should be remembered?

Ire marks a turning point, offering a post-signature era for malware defense—though not without risks.

5️⃣ A Pentesting AI Dethrones Humans on HackerOne

Summary : XBOW developed an automated pentesting tool that uses AI to detect real flaws, validated deterministically. It reached #1 on HackerOne’s US leaderboard—a first for a non-human actor.

Details:

• CTF methodology: Functions like a capture-the-flag player, planting canaries.

• Massive scope: 17,000 synthetic apps tested, 174 confirmed vulnerabilities, 22 real CVEs.

• Reliability: Unlike many LLMs, it doesn’t hallucinate—each bug validated logically.

• Critique: Naïve use of LLMs causes floods of false bug reports.

• Strategic impact: Shows a well-designed AI agent can surpass humans in rigor, volume, and consistency.

What should be remembered?

A new generation of AI audit tools is emerging, capable of large-scale scanning and validation, often outperforming human experts.

LIEUTENANT’S REPORT 🏆

Founded in 1976 in Canada, CGI is one of the world’s tech and strategy consulting giants. Present in over 40 countries with nearly 95,000 employees, CGI delivers end-to-end services : systems integration, audits, and cybersecurity.

Highlight :

In 2024, CGI was recognized as a global leader in cybersecurity consulting by IDC MarketScape.

CGI deploys end-to-end cybersecurity strategy : risk management, governance, secure engineering, incident response, and cyber intelligence.

As a true digital war commander, CGI blends human expertise and advanced technologies to anticipate threats, architect strong defenses, and neutralize attacks.

Its international presence and insights-driven culture make it an indispensable ally in a relentless cyberwar.

CYBERTRIVIA - DID YOU KNOW? 🤔

In July 2025, a bug in an old Windows API exposed over 20 million endpoints to an authentication bypass via a network printer !

Dubbed PrintSpoof Redux, it allowed attackers to use a crafted RPC request to gain SYSTEM rights on unpatched machines. 

Detected by a behavioral AI, it was exploited in targeted attacks before Microsoft released an emergency patch.

Even banal components—printers, drivers, forgotten protocols—can become critical entry points when neglected.

SURVEY 📊

👉️ Let me know if you need further adjustments !