AI News: The Day's Highlights

June 15, 2026 · 3 min read

Perplexity

The most notable recent news in this topic is not about the launch of another model, but about the strengthening trend of AI abandonment due to disappointment with business impact and risks of cost overload. According to the publication, by 2027, more than 40% of AI projects may be closed due to inefficiency, high costs, and ethical/management issues. [1]

Event Date: 2026; the publication identifies this shift as a current trend in late 2025 – early 2026. This perspective is noticeably different from the usual "news of the day" about service launches: here, the focus is on a broader market rollback and a reassessment of AI implementation strategies. [1]

Sources:

Sidorin Lab, "The Trend of AI Abandonment in 2026: How it Started and Whether to..."

Gemini

In 2025, a 30-fold increase in confidential information leaks from Russian companies through publicly available AI services such as ChatGPT and Gemini was recorded. The main reason was the active use of these tools by employees for analyzing work documents, spreadsheets, and code without proper awareness of the risks. Approximately 60% of Russian organizations still lack formalized policies regulating AI use, creating critical vulnerabilities for businesses.

Incidents involving AI services have become more frequent and diverse. Research by Check Point Software Technologies showed that 54% of cybersecurity specialists encountered AI-related incidents, including unauthorized AI use, the application of AI-generated content in attacks (e.g., phishing), and the leakage of confidential data through AI services. It is noted that the pace of AI adoption outpaces readiness to ensure security.

The Adversa AI report "Top AI Security Incidents – 2025 Edition" states that AI-related security incidents have doubled compared to 2024, with generative AI featuring in 70% of cases. "Agent" AI systems pose a particular danger, causing the most serious disruptions, including cryptocurrency theft and API abuse. Experts emphasize that AI is becoming a new attack surface, and simple prompt injection is a new "zero-day vulnerability," causing 35% of real AI security incidents.

Sources:

IT Brew
Adversa AI
Keepnet Labs
Stealth Cloud
CNews
National Business
Cybernews
Financial One
Moskovskaya Pravda
Anti-Malware.ru

ChatGPT

In June 2024, Snowflake reported a significant security incident involving a customer data breach. The research firm Mandiant discovered that hackers, allegedly linked to the UNC5537 group, gained access to data from hundreds of Snowflake customers using stolen credentials. Approximately 165 customers were affected by this incident, including companies like Ticketmaster and LendingTree. Snowflake has notified the affected customers and strengthened its security measures, including the implementation of multi-factor authentication. (techcrunch.com)

Grok

In the first quarter of 2026, the number of serious disruptions in the operation of major AI platforms sharply increased: according to Downdetector analysis, the number of days with an abnormally high level of user complaints rose from 6 in the same period of 2025 to 51. This affected services such as ChatGPT, Claude, Gemini, and Microsoft Copilot, as well as related cloud platforms AWS and Azure.

The growth reflects the transition of AI from an experimental stage to critical infrastructure, where even short-term outages affect millions of users and business processes. Analysts attribute this to the increasing complexity of systems, growing load, and reliance on cloud components. The event covers the period from January to April 2026, with data published in mid-June.

Claude

On June 9, 2026, at 5:21 PM EST, Anthropic received an order from the U.S. government and disabled foreign access to its most advanced models, Claude Fable 5 and Mythos 5, citing national security. The company stated that it considered this a misunderstanding and was working to restore access as soon as possible.

This radical decision reflects the growing pressure from the state on AI developers. The restriction on the use of advanced models applies to foreigners regardless of their location, both inside and outside the U.S. The decision will affect not only international users but also the overseas operations of American companies that have already integrated these models into their operations.

Concurrently, 2026 is marked by a surge in security incidents, with AI-driven attacks increasing by 89 percent. The Anthropic episode shows that threats now come not only from cybercriminals but also from government regulation.

Sources:

The Hacker News

Sources:

Blog