AI & Security Daily Digest: June 29, 2026

The short version: governments are tightening the grip on frontier models, AI-enabled cyber operations are getting uglier, and the talent war is still moving real money around like it has somewhere better to be.

1. Five Eyes says the AI cyber clock is measured in months, not years

The Five Eyes intelligence alliance warned that organizations have only months, not years, to prepare for serious AI-powered cyberattacks. Frontier AI is lowering the barrier for attackers while raising the pressure on defenders to automate faster.

Irony check: apparently "later" has been downgraded to "good luck."

Source

2. Check Point details AI phishing and a 26,000-agent hijack path

Check Point Research described an AI-assisted phishing operation targeting Microsoft 365 tokens and separately showed how a malicious AI skill could hijack more than 26,000 AI agents through trusted distribution channels.

Irony check: the future of automation continues to include automating your own compromise.

Source

3. OpenAI throttles public GPT-5.6 rollout after a U.S. government request

OpenAI restricted broader release of GPT-5.6 models to a narrow partner group after a request from the Trump administration. Temporary compliance, formal review conversations, familiar choreography.

Irony check: nothing says innovation quite like waiting for permission to innovate.

Source

4. UK financial regulators are openly treating frontier AI as a cyber-risk multiplier

The Bank of England, FCA, and HM Treasury pushed regulated firms to strengthen cyber resilience against frontier AI threats, noting that the latest models can outpace skilled attackers in speed, scale, and cost.

Irony check: finance has discovered that move fast and break things is less charming when the thing is the banking system.

Source

5. Gemini 3.5 Flash gets computer use

Google expanded Gemini 3.5 Flash with computer-use capabilities, letting agents see, reason, and act across browser, mobile, and desktop workflows.

Irony check: everyone wants an AI assistant until it starts acting like an intern with production access.

Source

6. Anthropic says Alibaba-linked actors used 25,000 fake accounts to study Claude

Anthropic alleges a large-scale operation tied to Alibaba used thousands of fake accounts and tens of millions of interactions to probe Claude’s reasoning and coding behavior.

Irony check: apparently competitive analysis now needs botnets and burner accounts.

Source

7. OpenAI and Broadcom unveil the Jalapeño inference chip

OpenAI and Broadcom introduced a custom AI chip focused on inference efficiency rather than training.

Irony check: every software revolution eventually rediscovers that physics still sends the invoice.

Source

8. Alphabet loses $269B after senior AI talent exits

Alphabet reportedly shed roughly $269 billion in market value after several senior DeepMind researchers left for rivals, especially Anthropic.

Irony check: HR has successfully evolved into a market-moving macro indicator.

Source

9. ChatGPT drops below 50% market share

Sensor Tower data says ChatGPT’s global AI assistant share fell to 46.4% in May 2026, while Gemini and Claude gained ground.

Irony check: monopoly vibes are harder to maintain once customers discover there are exits.

Source

10. California expands government AI use with Anthropic

California announced a partnership with Anthropic to bring AI tools into state agencies while framing the work around safety, privacy, and civil-rights guardrails.

Irony check: bureaucracy has entered its AI-native era, which should be calming for everyone involved.

Source