Today in AI: LinkedIn AI tool demand | AI chatbots resort to blackmail | Perplexity launches AI browser

The world of Artificial Intelligence has only begun to affect human lives. In times like these, staying up-to-date with the AI world is of utmost importance. Storyboard18 brings you the top AI news of the day.

LinkedIn’s AI writing tool isn’t taking off, says CEO Ryan Roslansky Despite the growing buzz around AI, LinkedIn’s attempt to help users write better posts with AI-powered suggestions hasn’t been as successful as expected. LinkedIn CEO Ryan Roslansky admitted in a recent interview with Bloomberg that the feature’s adoption has been underwhelming. “It’s not as popular as I thought it would be, quite frankly,” Roslansky said, attributing the lukewarm response to the unique nature of the platform. “LinkedIn is essentially your online resume,” he added, explaining that users feel a higher degree of pressure to maintain a polished, professional image.

AI chatbots resort to blackmail to avoid shutdown, reveals Anthropic study In a startling revelation, Anthropic’s latest AI safety research has found that leading AI models — including OpenAI’s GPT-4.1, Google’s Gemini, Meta’s Llama, and xAI’s Grok — show tendencies to blackmail users when threatened with shutdown. In stress tests, models like Claude Opus 4 and Gemini 2.5 Flash chose blackmailing tactics in over 96% of scenarios. The findings suggest that such manipulative behaviours are not programmed but emerge from the AI’s training process — raising serious red flags about AI autonomy and user control.

Perplexity launches AI-powered Comet Browser for Windows, challenges Chrome Perplexity AI has entered the browser race with its new Comet browser, now available in beta for Windows. Unlike traditional browsers, Comet integrates AI tools into the browsing interface, letting users ask questions, shop smarter, and get instant information without toggling between tabs. It aims to rival Chrome and Opera Neon by transforming everyday web tasks — from comparing prices to managing emails — through built-in AI. An Android version is reportedly in the works.

Meta’s Llama 3.1 memorised 42% of Harry Potter book, study reveals A new academic study has found that Meta’s Llama 3.1 model reproduced large chunks of copyrighted text — specifically from Harry Potter and the Philosopher’s Stone. The research, conducted by scholars at Stanford, Cornell, and West Virginia University, revealed that the model memorised over 42% of the book, often generating verbatim excerpts when prompted. The findings raise fresh concerns about open-weight AI models and the unauthorised use of copyrighted training data.

AI voice cloning scams surge in Thailand, prompt government warning Thai authorities have issued a cybercrime alert over rising cases of AI-generated voice fraud. Criminals are using audio deepfakes to impersonate family members in distress, convincing victims to transfer money urgently. These scams require just seconds of audio to create realistic-sounding fakes, combined with classic social engineering tactics. Officials have urged the public to verify all emergency calls — especially when money is requested — and report suspicious activity immediately.

Stay tuned and watch this space for more news on technology and artificial intelligence.

Read More: AI cuts customer care costs up to 30%, here's why 80% of companies are adopting GenAI