Angela Lipps, a 50-year-old Tennessee woman, was arrested at gunpoint after AI facial recognition software matched her face to a blurry surveillance image from a North Dakota bank fraud case — a state she says she had never visited. Fargo police relied on a neighboring city's facial recognition system, which flagged Lipps as a "potential suspect," and conducted no further investigation before arresting her. She spent 108 days in jail. On Christmas Eve, her lawyers showed police that her bank records placed her in Tennessee — buying cigarettes and depositing Social Security checks — at the time of the alleged crimes. Her case was dismissed. She had lost her house, her car, and her dog. CNN published the story on March 29; her attorneys are now exploring civil rights claims.
Why it matters: AI-generated outputs are increasingly used as evidence in hiring, fraud detection, and security systems. If your company uses AI, please remember that it's worth a hard look at your human review process.
Sources: CNN · Tom's Hardware
Mistral released Voxtral TTS on March 26, its first text-to-speech model, and it's making the full weights available for free download on Hugging Face. The model supports nine languages, runs in real time with a 90ms time-to-first-audio latency, and can voice-clone from as little as 3 seconds of reference audio. At 4 billion parameters, it's compact enough to run on a laptop or even a smartphone. For teams that don't want to self-host, the API costs $0.016 per 1,000 characters. Mistral claims it outperforms ElevenLabs on quality benchmarks.
Why it matters: Until now, production-quality text-to-speech meant paying ElevenLabs, OpenAI, or similar. Mistral just changed that math by open-sourcing a model that runs locally, supports multiple languages, and doesn't require sending audio data to a third party. For businesses building customer-facing voice features, internal training tools, or multilingual content, this is worth evaluating — especially for teams with privacy or cost concerns about hosted APIs.
Sources: TechCrunch · VentureBeat · Mistral
OpenAI rolled out updated integrations for Notion, Box, Linear, and Dropbox in ChatGPT over the weekend, adding write capabilities for the first time. Previously, ChatGPT could pull content from these apps; now it can create pages, update tasks, and push changes back directly. Enterprise and Business users need to reconnect their apps and review the new action permissions in workspace settings; standard users need to reconnect to unlock the update. Notion and Linear are also being added as synced connectors for Pro users.
Why it matters: Read-only integrations are useful. Read-write integrations are transformative. If your team uses any of these tools, ChatGPT can now do real work in your systems rather than just summarize them. Worth spending 20 minutes reconnecting these apps and testing — updating a Linear issue from a ChatGPT conversation, or drafting and saving a Notion page without leaving chat, are the kinds of things that add up to significant time savings.
Sources: OpenAI Release Notes · OpenAI Community
Claude Code 2.1.87 (released Sunday) fixes a message delivery bug in Cowork Dispatch that was causing notifications not to arrive. If you've noticed missed messages in Dispatch this weekend, updating should resolve it. (Releasebot)
Meta published leaked internal targets showing it wants 50% of new code written by AI by the end of 2026, according to documents reviewed by The Week. Zuckerberg has been framing Meta's shift to being "AI-native" publicly for months; these documents reveal the specific engineering benchmarks he's driving toward. (The Week)
Forward this briefing to a colleague who wants to stay current on AI without the jargon. They can subscribe at your.biggestgoal.ai to get it daily.
Curated by Biggest Goal · March 30, 2026
Angela Lipps, a 50-year-old Tennessee woman, was arrested at gunpoint after AI facial recognition software matched her face to a blurry surveillance image from a North Dakota bank fraud case — a state she says she had never visited. Fargo police relied on a neighboring city's facial recognition system, which flagged Lipps as a "potential suspect," and conducted no further investigation before arresting her. She spent 108 days in jail. On Christmas Eve, her lawyers showed police that her bank records placed her in Tennessee — buying cigarettes and depositing Social Security checks — at the time of the alleged crimes. Her case was dismissed. She had lost her house, her car, and her dog. CNN published the story on March 29; her attorneys are now exploring civil rights claims.
Why it matters: AI-generated outputs are increasingly used as evidence in hiring, fraud detection, and security systems. If your company uses AI, please remember that it's worth a hard look at your human review process.
Sources: CNN · Tom's Hardware
Mistral released Voxtral TTS on March 26, its first text-to-speech model, and it's making the full weights available for free download on Hugging Face. The model supports nine languages, runs in real time with a 90ms time-to-first-audio latency, and can voice-clone from as little as 3 seconds of reference audio. At 4 billion parameters, it's compact enough to run on a laptop or even a smartphone. For teams that don't want to self-host, the API costs $0.016 per 1,000 characters. Mistral claims it outperforms ElevenLabs on quality benchmarks.
Why it matters: Until now, production-quality text-to-speech meant paying ElevenLabs, OpenAI, or similar. Mistral just changed that math by open-sourcing a model that runs locally, supports multiple languages, and doesn't require sending audio data to a third party. For businesses building customer-facing voice features, internal training tools, or multilingual content, this is worth evaluating — especially for teams with privacy or cost concerns about hosted APIs.
Sources: TechCrunch · VentureBeat · Mistral
OpenAI rolled out updated integrations for Notion, Box, Linear, and Dropbox in ChatGPT over the weekend, adding write capabilities for the first time. Previously, ChatGPT could pull content from these apps; now it can create pages, update tasks, and push changes back directly. Enterprise and Business users need to reconnect their apps and review the new action permissions in workspace settings; standard users need to reconnect to unlock the update. Notion and Linear are also being added as synced connectors for Pro users.
Why it matters: Read-only integrations are useful. Read-write integrations are transformative. If your team uses any of these tools, ChatGPT can now do real work in your systems rather than just summarize them. Worth spending 20 minutes reconnecting these apps and testing — updating a Linear issue from a ChatGPT conversation, or drafting and saving a Notion page without leaving chat, are the kinds of things that add up to significant time savings.
Sources: OpenAI Release Notes · OpenAI Community
Claude Code 2.1.87 (released Sunday) fixes a message delivery bug in Cowork Dispatch that was causing notifications not to arrive. If you've noticed missed messages in Dispatch this weekend, updating should resolve it. (Releasebot)
Meta published leaked internal targets showing it wants 50% of new code written by AI by the end of 2026, according to documents reviewed by The Week. Zuckerberg has been framing Meta's shift to being "AI-native" publicly for months; these documents reveal the specific engineering benchmarks he's driving toward. (The Week)
Forward this briefing to a colleague who wants to stay current on AI without the jargon. They can subscribe at your.biggestgoal.ai to get it daily.
Curated by Biggest Goal · March 30, 2026