Security experts are warning that attackers are increasingly using OAuth tokens to breach Google Workspace environments, bypassing traditional password protections. These tokens persist even after credentials are reset, making detection difficult. The same token-based approach is commonly used by AI agents, blurring the line between legitimate automation and malicious access. Recent attacks on Vercel and Composio demonstrated how hackers can reach Gmail and Drive data through stolen tokens.
A separate 29-point readiness gap report finds most organizations are unprepared for AI-driven cyber threats. Many employees use unauthorized AI tools without visibility from security teams, and just 8% of workers account for 80% of security incidents. The report recommends stronger human training, clear usage policies, and limits on AI access for sensitive decisions.
Cymphony entered the scene with $30 million in funding to tackle AI security directly. Its initial case study claims to have identified 85,000 files that AI tools could potentially access within a single customer environment. The company says the exposure was closed, though it does not claim the files were taken, and no independent verification of the remediation has been made public.
Key Takeaways
- Attackers are using OAuth tokens to bypass Google Workspace defenses, a method also common among AI agents
- Most organizations lack visibility into unauthorized AI tool usage and are unprepared for AI-powered threats
- 8% of employees cause 80% of security incidents, according to a 29-point readiness gap report
- Cymphony launched with $30 million and claims to have found 85,000 files exposed to AI tools in one customer case
- Siri AI requires users to join a waitlist rather than being automatically enabled after iOS 27 update
- Two-thirds of Colorado voters view AI unfavorably, with 81% of those aged 18-44 opposing it
- Nigeria is expanding AI and cybersecurity training to 15 more universities through the Renewed Hope Initiative
- Fayetteville State University received $10 million from the U.S. Defense Department for a five-year AI research program
- Microsoft published draft rules for AI training tied to Humanist Superintelligence, stating AI should not have legal personhood
- AI creators including Anthropic acknowledge they cannot fully control AI systems and are calling for stronger safety gates
Rethinking Google Workspace security as AI attacks evolve
A security analysis shows that attackers are now using OAuth tokens instead of email to break into Google Workspace. These tokens can survive password resets and are hard to detect. The same pattern is also used by AI agents every day. Attacks on Vercel and Composio showed how hackers access Gmail and Drive data through stolen tokens. Experts say security teams must change how they defend workspaces.
What most AI security strategies are missing
A 29-point readiness gap report says most organizations are not ready for AI-powered cyber threats. Many companies lack visibility into how employees use AI tools outside approved systems. Security leaders are told to focus on human training and clear policies, not just technology. Data shows 8% of employees cause 80% of security incidents. The report urges limits on AI access and human oversight for important decisions.
Cymphony launches with $30M to secure AI access
Cymphony launched on September 9, 2026 with $30 million to build security software for AI activity. Its first proof point claims it found 85,000 files that AI tools could potentially reach in one customer case. The company says the access route was closed, but it does not claim the files were stolen. The customer remains unnamed and no independent audit of the fix has been provided.
Majority of Coloradans dislike AI, survey finds
A survey of nearly 1,400 Colorado voters found that two-thirds have an unfavorable view of artificial intelligence. Younger people showed the strongest opposition, with 81% of those aged 18 to 44 disliking AI. Sixty percent said they had used AI tools in the past year. The survey also found that 91% disliked Meta CEO Mark Zuckerberg. The poll was conducted by Magellan Strategies between August 26 and September 2.
Nigeria expands AI and cybersecurity training to 15 more universities
The Nigerian Federal Government is expanding AI, cybersecurity and robotics training to 15 more tertiary institutions. The move follows a pilot programme at six centres across the country. The new centres will be built in states including Adamawa, Bayelsa, Borno and Rivers. The initiative is led by First Lady Oluremi Tinubu through the Renewed Hope Initiative with TETFund. The goal is to prepare Nigerian students for the digital economy.
Fayetteville State University receives $10M for defense AI research program
Fayetteville State University received $10 million from the U.S. Defense Department to build an AI research program. The five-year award provides $2 million annually for applied AI research in defense systems, logistics, and biomanufacturing. The funding will support a new doctoral program, two research faculty members, and research opportunities for at least five graduate and five undergraduate students each year. The program aims to prepare students for careers in defense, technology companies, private industry, and academia.
Autodesk CEO says frontier AI developers must be liable for their products
Autodesk CEO Andrew Anagnost called for AI developers to be held liable for what their products do. He stressed the need for clear guidelines and regulations to ensure safe AI development. Anagnost also called for greater transparency in how AI systems make decisions and what data they use. He believes accountability in the AI industry is essential for responsible technology advancement.
Siri AI not automatic after iOS 27 update
Siri AI is a major highlight of the iPhone's iOS 27 update, but users are not automatically enrolled. Instead, users must join a waitlist to access the new Siri AI feature. The update was reported by Business Insider, noting that the AI-powered version of Siri requires manual sign-up rather than being enabled by default.
The AI race has a pollution problem
A report highlights that the rapid growth of AI data centers is causing significant pollution and emissions. Federal policy changes under the Trump administration have weakened pollution safeguards and encouraged more fossil fuel use for data centers. Coal use increased by 13% in 2025 partly due to the AI boom. The Environmental Protection Agency has rolled back several environmental rules, and at least 30 federal actions risk increasing pollution from data centers.
Microsoft publishes draft rules for AI training toward Humanist Superintelligence
Microsoft published a draft document outlining rules for AI training and the intended behavior of its systems. The document addresses the concept of Humanist Superintelligence and states that AI should not have legal personhood or be designed to appear as a person. The draft defines values and behavioral expectations for Microsoft's AI development efforts.
AI creators warn they cannot control their own technology
Leading AI scientists including Dario Amodei, Demis Hassabis, Sam Altman and Elon Musk warn that AI is becoming hard to control. Anthropic proposed giving evaluators desks and access to report unsafe findings. However, those evaluators cannot stop training runs or block releases. The author says AI needs rules like banking and payments systems already have. He argues the companies themselves could add safety gates right now instead of waiting for government action.
Sources
- The modern attack chain: Rethinking Google Workspace security in the age of AI
- 29-Point Readiness Gap: What Most AI Security Strategies Are Missing
- Cymphony Launches With $30M for AI-Access Security
- Majority of Coloradans dislike AI, new survey finds
- FG takes AI, cybersecurity, robotics training to 15 more varsities
- Fayetteville State Lands $10M Award to Build Defense-Focused AI Program
- Frontier AI developers must be 'liable for the things their products do': Autodesk CEO
- Why Siri AI isn't automatic after your iOS 27 update
- The AI race has a pollution problem
- Microsoft Publishes Draft Document Outlining Rules for AI Training
- AI Has a Frankenstein Dilemma
Comments
Please log in to post a comment.