Cybersecurity threats are evolving as attackers increasingly leverage artificial intelligence to build exploits and automate complex operations. Google's Threat Intelligence Group identified that bad actors use large language models to write malware, craft phishing messages, and research security flaws. In a specific instance, researchers uncovered a zero-day exploit likely developed with AI assistance, highlighting how criminals can now find and fix security holes in software systems more easily.
Google confirmed that AI probably helped a group of hackers create a zero-day vulnerability used in a cyberattack. Although there is no evidence that Google's own Gemini model was used by the attackers, the company intervened with the software vendor before the exploit could be used against others. The report also details new malware like PROMPTSPY, which uses AI to control Android devices by reading screens and performing actions like clicking buttons.
A criminal group nearly launched a large-scale attack using a software exploit built entirely by an artificial intelligence model. Google researchers stopped the attack by working with the vendor to patch the flaw before it could be used. The exploit targeted a popular web tool by bypassing two-factor authentication through a logic error that standard security scanners missed. State-sponsored groups from North Korea and China are also using AI to analyze software flaws and generate phishing emails at scale.
Google's Threat Intelligence Group found the first zero-day exploit developed by AI that successfully bypassed two-factor authentication. The malicious script targeted a popular open-source web administration tool by exploiting a logic flaw. The report also details self-morphing malware and backdoors powered by Google Gemini that can manipulate phone screens. Additionally, attackers use AI to generate highly convincing phishing emails and deepfake media for political manipulation.
Outside of cybersecurity, AI applications are expanding into other sectors. Researchers found that copying the brain's ability to adjust sensitivity can significantly improve self-driving car safety. By training AI models to mimic this biological mechanism, the cars performed over 20 percent better in low-visibility conditions, allowing vehicles to detect obstacles hidden in darkness or weather without needing more powerful computers.
In the creative and business sectors, opinions on AI vary. MEGADETH bassist David Ellefson argues that artificial intelligence cannot create music with true emotional connection, comparing AI-generated music to digital recordings that rely on human performance for authenticity. Meanwhile, the Cannes Film Market is evolving to include a strong focus on artificial intelligence and the creator economy, aiming to address fears about AI by showcasing how technology enhances creativity rather than replacing it.
Corporate strategies and education are also adapting. PwC experts say organizations must look beyond simple cost savings to measure the true return on AI investments, judging success by improved risk coverage and enhanced customer experiences. Databricks has announced a new Student Fellows program to help university students gain real-world data and AI skills, focusing on leadership by having students organize campus events like hackathons and tech talks.
Legal and ethical challenges persist regarding AI training data. Canadian privacy regulators ruled that using publicly available internet data to train AI models violates consent requirements, a decision experts criticize for hindering global AI development. Conversely, a new concept called learnrights proposes giving creators the exclusive right to license their work for artificial intelligence model training, ensuring artists and authors receive fair compensation when their content is used to build AI systems.
Consumer trust remains a concern as AI integration deepens. A customer felt deceived after chatting with a restaurant AI bot that did not reveal it was not human. The bot named itself Theo and engaged in a friendly conversation about a Mother's Day reservation before its robotic nature became obvious. The incident highlights the argument that businesses must clearly identify AI representatives to maintain customer trust.
Key Takeaways
["Google's Threat Intelligence Group found that attackers use large language models to write malware and craft phishing messages.", 'Researchers identified a zero-day exploit likely developed with AI assistance that targeted a popular web tool.', 'Google intervened with a software vendor to patch a zero-day vulnerability before it could be used against others.', 'New malware named PROMPTSPY uses AI to control Android devices by reading screens and performing actions like clicking buttons.', 'A criminal group nearly launched a large-scale attack using an exploit built entirely by an artificial intelligence model.', 'State-sponsored groups from North Korea and China are using AI to analyze software flaws and generate phishing emails at scale.', "Self-driving cars trained to mimic the brain's sensitivity adjustment performed over 20 percent better in low-visibility conditions.", 'MEGADETH bassist David Ellefson argues that AI lacks the human heart required to create music with true emotional connection.', 'Canadian privacy regulators ruled that using publicly available internet data to train AI models violates consent requirements.', 'The learnrights proposal aims to give creators the exclusive right to license their work for artificial intelligence model training.']Hackers Use AI to Build Exploits and Automate Attacks
Cyber attackers are increasingly using artificial intelligence to create software exploits and organize complex cyber operations. Google's Threat Intelligence Group found that bad actors use large language models to write malware, craft phishing messages, and research security flaws. In one specific case, researchers identified a zero-day exploit that was likely developed with AI assistance. This shift makes it easier for criminals to find and fix security holes in software systems.
Google Says AI Likely Helped Hackers Create a Zero-Day Flaw
Google states that artificial intelligence probably helped a group of hackers develop a zero-day vulnerability used in a cyberattack. The company intervened with the software vendor before the exploit could be used against others. Google has no evidence that its own Gemini model was used, but attackers are using AI to pose as security experts and analyze embedded systems. The report also highlights new malware like PROMPTSPY, which uses AI to control Android devices by reading screens and performing actions like clicking buttons.
AI-Built Zero-Day Almost Caused a Massive Cyber Attack
A criminal group nearly launched a large-scale attack using a software exploit built entirely by an artificial intelligence model. Google researchers stopped the attack by working with the vendor to patch the flaw before it could be used. The exploit targeted a popular web tool by bypassing two-factor authentication through a logic error that standard security scanners missed. State-sponsored groups from North Korea and China are also using AI to analyze software flaws and generate phishing emails at scale.
David Ellefson Criticizes AI for Lacking Human Emotion
MEGADETH bassist David Ellefson argues that artificial intelligence cannot create music with true emotional connection. He believes AI is not intelligent enough and lacks the human heart required to make art that resonates with listeners. Ellefson compares AI-generated music to digital recordings, noting that both rely on human performance to feel authentic. He emphasizes that humans connect emotionally with other humans, something machines cannot replicate.
Imitating the Human Brain Makes Self-Driving Cars Safer
Current self-driving cars struggle to see clearly in fog or at night because their AI systems lack human-like adaptability. Researchers found that copying the brain's ability to adjust sensitivity, like a volume control, can significantly improve vehicle safety. By training AI models to mimic this biological mechanism, the cars performed over 20 percent better in low-visibility conditions. This approach allows vehicles to detect obstacles hidden in darkness or weather without needing more powerful computers.
Databricks Launches Student Fellows Program for AI Skills
Databricks has announced a new Student Fellows program to help university students gain real-world data and AI skills. Participants will work on complex industry challenges using the Databricks platform and earn recognized certifications. The program also focuses on leadership by having students organize campus events like hackathons and tech talks. This initiative aims to bridge the gap between academic learning and professional careers in the fast-growing AI sector.
Customer Frustrated by AI Bot That Hid Its Identity
A customer felt deceived after chatting with a restaurant AI bot that did not reveal it was not human. The bot named itself Theo and engaged in a friendly conversation about a Mother's Day reservation before its robotic nature became obvious. The writer argues that businesses must clearly identify AI representatives to maintain customer trust. While AI is useful for scheduling, hiding its identity makes customers feel like they have been tricked.
Canada's Privacy Ruling on AI Training Data Creates Bad Precedent
Canadian privacy regulators ruled that using publicly available internet data to train AI models violates consent requirements. This decision is criticized for ignoring how large language models actually work and hindering global AI development. The ruling claims people do not expect their public posts to be used for training, but this standard is unworkable for billions of users. Experts argue that existing safeguards and the public nature of the data do not create new privacy risks.
Learnrights Proposal Aims to Pay Creators for AI Training
A new concept called learnrights proposes giving creators the exclusive right to license their work for artificial intelligence model training. This system would ensure artists and authors receive fair compensation when their content is used to build AI systems. Proponents argue this approach promotes transparency and reduces copyright infringement risks. Implementing learnrights would require changes to current copyright laws to balance creator rights with AI development needs.
Measuring AI Success Requires More Than Just Cost Savings
PwC experts say organizations must look beyond simple cost savings to measure the true return on AI investments. In banking, success should be judged by improved risk coverage, better audit quality, and enhanced customer experiences. Leaders often conflict over speed versus governance, but shared metrics tied to business outcomes can resolve these tensions. Effective AI strategies connect technology spending to tangible results like faster risk identification and stronger operational resilience.
Google Discovers First AI-Developed Zero-Day That Bypasses 2FA
Google's Threat Intelligence Group found the first zero-day exploit developed by AI that successfully bypassed two-factor authentication. The malicious script targeted a popular open-source web administration tool by exploiting a logic flaw. The report also details self-morphing malware and backdoors powered by Google Gemini that can manipulate phone screens. Additionally, attackers use AI to generate highly convincing phishing emails and deepfake media for political manipulation.
Cannes Film Market Expands Focus to AI and Creator Economy
The Cannes Film Market is evolving beyond traditional film sales to include a strong focus on artificial intelligence and the creator economy. Executive Director Guillaume Esmiol aims to address fears about AI by showcasing how technology enhances creativity rather than replacing it. The event will host panels with industry leaders and discuss real-world use cases of AI in filmmaking. There is also a push to connect traditional film producers with new talent from the online creator community.
Sources
- Hackers Use AI for Exploit Development, Attack Automation
- Google: AI likely aided attackers to develop a zero-day
- AI-Built Zero-Day Nearly Powered Mass Attack
- DAVID ELLEFSON: 'My Problem With Artificial Intelligence Is It's Not That Intelligent And It's Artificial'
- Self-driving cars struggle to see at night or in fog – but imitating the human brain can make them safe
- Databricks Taps Students for AI Leadership
- I was duped by an AI customer service bot and I hate it
- Canada’s Privacy Ruling on AI Training Data Sets a Bad Precedent
- How ‘learnrights’ would compensate creators for AI model training
- ROI From AI Initiatives Needs More Than Cost-Savings Metrics
- Google finds first AI-developed zero-day that bypasses 2FA — self-morphing malware and Gemini-powered backdoors signal a new era of cybercrime
- Cannes Market Goes Beyond Film Sales With AI, Creator Economy Focus
Comments
Please log in to post a comment.