OpenAI Begins Offering Cyber-Specialized Model "GPT-5.5-Cyber" via Trusted Access

On May 7, OpenAI expanded its Trusted Access for Cyber program and began offering the new model "GPT-5.5" along with a cybersecurity-specialized version, "GPT-5.5-Cyber." These models are made available to vetted security defenders, with the aim of accelerating vulnerability research and protecting critical infrastructure.

GPT-5.5-Cyber is a model optimized specifically for security research and is positioned as a tool for defenders rather than attackers. It's fair to say the trend of AI being put to practical use on the front lines of cybersecurity is accelerating further. The fact that OpenAI has established a vetting framework called "Trusted Access" draws attention as a sign of its intent to control the distribution of highly capable models.

Cloudflare Announces Layoff of 1,100 Employees, Citing AI

Cloudflare, a major CDN and cloud security company, revealed alongside its Q1 earnings announcement that it would cut 1,100 employees. The transformation of business processes driven by AI was cited as the reason.

Large tech companies are increasingly carrying out massive layoffs citing AI adoption. On the same day, Coinbase also announced a reduction of about 14% of its workforce, attributing it to changes in ways of working brought about by AI — making it clear that AI is shifting from a mere business efficiency tool to a factor that is reshaping the very human structure of organizations.

Trump Administration Shifts on AI Regulation — Considering Federal Oversight via Executive Order

According to a WIRED report, the Trump administration is considering issuing an executive order to establish federal-level pre-oversight of AI models. The move signals a shift in the administration's stance, which had previously been reluctant to regulate AI.

While the details have yet to be finalized, discussions are reportedly progressing toward putting some form of federal-level oversight in place for the development and release of new AI models. The debate over how AI should be regulated appears to be increasingly recognized as an important issue that transcends party lines.

EU Bans Non-Consensual AI-Generated Sexual Imagery — Agreement Reached on AI Act Amendment

The EU has reached a provisional agreement on amendments to the AI Act. While the amendments include some regulatory easing, they also newly establish provisions explicitly banning the generation and use of sexual imagery with AI without consent.

Damage from sexual deepfake imagery is becoming increasingly severe worldwide, and the EU's move to address it through legislation is highly significant. The EU's stance of seeking a balance between deregulation and tighter regulation is also reflected in this amendment.

Security Concerns Raised Over Microsoft's AI Feature — Data Leakage Feared Even with "Ironclad" Encryption

According to an ITmedia report, security researchers have pointed out that a Microsoft AI feature has a weakness "not in the data itself, but in the process of the exchanges."

Even if the encryption itself is robust, possible leakage paths through the communication process and metadata have been identified. This has underscored the limitations of an approach that relies on encryption alone in designing AI features.

Writers Turn to Extreme Measures to Prove They Are "Not Using AI"

A Wall Street Journal report has shed light on the reality that professional writers are beginning to take extreme measures to prove they are not using AI. False positives from AI detection tools keep occurring, and this is a way to deal with the risk of their own writing being flagged as "AI-generated."

Various approaches are being attempted, including saving the entire history of the writing process, obtaining third-party verification, and deliberately maintaining a writing style that AI struggles to produce. As AI-generated content proliferates, proving that "a human wrote this" is emerging as a new challenge.

---