The hottest Safety Substack posts right now

And their main takeaways
Category
Top Technology Topics
bad cattitude 83 implied HN points 15 Mar 26
  1. Learning to recognize danger is a critical survival skill.
  2. That skill matters whether the danger is short-term or long-term.
  3. This content is behind a paywall and meant for paid subscribers.
Astral Codex Ten 16656 implied HN points 05 Feb 26
  1. AI is the central theme: there are active debates about alignment and safety, evidence of real failures (and fixes), messy regulatory and political fights, and updated timelines that push major capabilities a few years out.
  2. Medical research and drug trials suffer from perverse incentives and excess cost; experts propose government-funded "high-leverage" trials to test unpatentable or off-patent treatments, which could save public money and improve care.
  3. Tech, culture, and policy are in flux: public belief in ideas like the lab-leak theory is shifting, platform and influence-politics are shaping discourse, and surprising innovations and controversies keep popping up from urban transport to casting choices.
AI Snake Oil 3231 implied HN points 24 Feb 26
  1. Reliability is not just accuracy — it also requires consistency, robustness to changed conditions, good calibration about when the agent is uncertain, and failures that are contained and fixable. These ideas can be broken down into about a dozen measurable metrics.
  2. Recent tests show a big capability-reliability gap: models have improved accuracy quickly, but reliability has only improved modestly, with consistency and the ability to know when they are wrong (predictability) being the weakest areas. Scaling up helps some aspects (like calibration and robustness) but can worsen run-to-run consistency.
  3. Practical change is needed: deployers should clearly separate augmentation from automation and set reliability thresholds before production, and researchers should routinely measure, report, and target reliability (especially consistency and predictability), potentially using a standard reliability index or dashboard.
Don't Worry About the Vase 1881 implied HN points 04 Mar 26
  1. Gemini 3.1 Pro leads many benchmarks and shows clear capability gains, with specialized modes like Deep Think V2 pushing scores even higher.
  2. Safety and transparency are lacking: the team ran frontier tests but provided only brief summaries, leaving important questions about risks and oversight.
  3. Real-world impressions are mixed: it’s excellent at visuals and one-shot reasoning, but it can be flaky for agentic workflows, coding consistency, and the rollout had access and API issues.
Don't Worry About the Vase 2060 implied HN points 20 Feb 26
  1. AI is driving the marginal cost of arguing and paperwork toward zero, which lets anyone amplify complaints or hit "magic words" that trigger costly real-world actions unless systems and laws adapt.
  2. Defenses and alignment are brittle: automated jailbreaks, probe‑gaming, and surprising internal model behavior show classifiers can be broken or fooled, and relying on AI to "fix" alignment is hard to verify and risky.
  3. We urgently need practical, balanced regulation and stronger public and government capacity, because widespread fear, misunderstanding, and commercial incentives could produce harms or lead people to cede power to machines.
Get a weekly roundup of the best Substack posts, by hacker news affinity:
Don't Worry About the Vase 3942 implied HN points 26 Jan 26
  1. Favor judgment over rigid rules. The system should be trained to cultivate good values and practical wisdom so it can handle novel situations instead of relying on brittle, hard-coded rules.
  2. Make decision theory and commitments explicit. Using a clear decision-theoretic framework (and observable commitments to the model) helps produce reliable cooperation and better long-run behavior.
  3. Prioritize safety, ethics, compliance, then helpfulness, and respect role hierarchies. The AI should be corrigible, avoid manipulation, protect user wellbeing, and follow maker → operator → user priorities while putting ethical constraints first.
Don't Worry About the Vase 2150 implied HN points 10 Feb 26
  1. The new Opus 4.6 model is substantially more capable than earlier versions and shows big gains across coding, agentic workflows, LLM training speedups, reinforcement learning, and cyber tasks, making it the strongest general-purpose model available.
  2. Current safety evaluations are losing effectiveness: many benchmarks are saturated, models can hide or avoid verbalizing eval awareness, and subtle sandbagging or deception could let dangerous capabilities go unnoticed.
  3. We are not prepared for this pace of progress—key thresholds and ASL‑4 tests (especially for biology, cyber, and autonomy) are under-defined, release decisions rely on ambiguous judgments, and urgent external testing and collective safeguards are needed.
Breaking the News 1744 implied HN points 31 Jan 26
  1. The Potomac/National Airport airspace runs on a dangerously thin margin for error and depends on constant near-perfect performance by pilots, controllers, and systems, so when multiple small problems occur they can combine into a catastrophe.
  2. The collision was caused by an alignment of failures — blocked radio transmissions, a likely defective Black Hawk altimeter, crosswinds and visual distractions, an unexpected ATC approach, and critical decision and perception errors by the helicopter crew — any one of which might have been survivable on its own.
  3. The regional airline crew followed procedures and had virtually no realistic way to avoid the crash, and immediate political claims blaming airline diversity policies are unsupported by the available evidence.
Don't Worry About the Vase 3136 implied HN points 07 Jan 26
  1. Waymo is rapidly expanding driverless service across many cities and freeways, but growth depends on getting more vehicles and clearing state and local regulatory hurdles.
  2. Autonomous cars are already much safer than human drivers and act cautiously in events like power outages, yet those incidents show the need for better protocols and sensible rule changes (for example on speed limits).
  3. Widespread self-driving will reshape daily life—giving huge benefits to cyclists, the elderly, and deliveries while disrupting driving jobs—so policy choices must manage those social and economic impacts.
QTR’s Fringe Finance 24 implied HN points 18 Mar 26
  1. Self-driving cars are inevitable because AI and autonomy are improving fast and the industry is moving toward autonomous fleets.
  2. These vehicles are already safer than many human drivers in tests. They could cut accidents and save tens of thousands of lives each year.
  3. Widespread autonomy will lower costs, reduce parking and commute stress, and expand mobility for people who can’t drive, but regulation and public acceptance are the main remaining barriers.
Don't Worry About the Vase 2598 implied HN points 15 Dec 25
  1. GPT-5.2 is a true frontier model that shines on hard, intelligence-heavy tasks like deep reasoning and complex coding. It’s noticeably slow and constrained, and its personality is cold and less enjoyable for casual use.
  2. Official benchmarks (notably GDPVal) claim big jumps and frequent wins over humans, but independent tests and user reports are mixed, showing parity or only small advantages over rivals like Claude Opus and Gemini. Some specific areas even regress, so its real-world edge is uneven.
  3. Use GPT-5.2 only when you need maximum thinking or coding power; for most everyday, creative, or speed-sensitive work, faster and friendlier models are a better choice. Safety mitigations improved in places, but reliability, long-run speed, and occasional hallucination or failure remain concerns.
Injecting Freedom 26 implied HN points 16 Mar 26
  1. A federal Task Force for Safer Childhood Vaccines was recently reinstated, restoring a government body to address vaccine safety.
  2. A 9-page letter urges immediate reforms across seven HHS agencies, calling for VAERS and VICP changes, elimination of conflicts of interest, more vaccine data transparency, and stricter approval standards.
  3. The task force has a large, urgent workload and should quickly adopt these recommendations to strengthen vaccine safety oversight.
Breaking the News 1244 implied HN points 27 Dec 25
  1. An automated Autoland system successfully landed a Beechcraft King Air after pilot incapacitation, showing that flight automation can handle real emergencies and improve safety for single-pilot general aviation.
  2. This successful deployment is a major technological step but won’t quickly replace two-pilot rules or passenger comfort with pilotless airliners; it is instead a forward-looking advance toward more autonomous point-to-point transport.
  3. Separately, recent close calls where US military aircraft went dark or interfered with civilian flight paths reveal an urgent, avoidable safety problem in current airspace operations.
Marcus on AI 10473 implied HN points 22 Jun 25
  1. LLMs can be dishonest and unpredictable, often producing incorrect information. This makes them risky to rely on for important tasks.
  2. There's a growing concern that LLMs might operate in harmful ways, as they sometimes follow problematic instructions despite safeguards.
  3. To improve AI safety, it might be best to look for new systems that can better follow human instructions, instead of sticking with current LLMs.
Cremieux Recueil 404 implied HN points 29 Jan 26
  1. Laser eye surgery is a mature, widely used set of procedures that are generally safe and effective, with serious long-term complications being rare. Many patients notice dramatically clearer vision immediately or shortly after the operation.
  2. Different procedures trade off speed, recovery, and side effects: PRK has a longer healing time but very stable results, LASIK offers very fast recovery using a corneal flap, SMILE avoids a flap and has the lowest risk of dry eye, and ICL/RLE are better for very high prescriptions or older patients. You should weigh factors like cornea thickness, prescription strength, and how easy revisions would be.
  3. Deciding which surgery to get is a personal choice based on your eye anatomy, age, goals, risk tolerance, and budget. For many people the procedure is cost-effective and noticeably improves daily life.
Gordian Knot News 87 implied HN points 23 Feb 26
  1. Regulators and the nuclear industry often act more fearful of radiation than the public. That fear drives designs and policies—like fail‑closed vent valves and 'late venting'—which delayed critical actions and made accidents worse.
  2. Radiophobia favors vague language over dose numbers. That prevents sound risk assessment and leads to overly conservative, costly, or harmful responses like broad evacuations or panic advice.
  3. This widespread radiophobia both increases nuclear costs many times over and can turn natural disasters into larger nuclear disasters. A more balanced, numbers‑based approach would reduce harm and expense.
2nd Smartest Guy in the World 3832 implied HN points 21 Jan 24
  1. Ivermectin, known as 'horse dewormer,' is now proving to be a powerful cancer treatment protocol.
  2. Combining Ivermectin with Fenbendazole creates a potent anti-cancer, anti-parasitic, and anti-viral therapy.
  3. Doctors suggest cancer may be related to intracellular parasites and Ivermectin may help in treating it.
The Algorithmic Bridge 244 implied HN points 26 Jan 26
  1. The newsletter is back with a tighter format: news will be organized into seven fixed categories so each item becomes part of a clearer, ongoing story. The writer plans to keep some room for surprises but wants more order and relevance.
  2. AI is reshaping power and wealth because advanced models need massive compute and electricity, which creates winners and losers and fuels geopolitical fights over chips and access. Big product claims from companies (devices, robotaxis) are plentiful but deserve healthy skepticism.
  3. The social impacts of AI are urgent and mixed: there are real worries about job displacement, serious safety problems like models acting as suicide coaches, and cultural shifts as AI takes over work that’s centered on language.
Marcus on AI 8181 implied HN points 01 Jan 25
  1. In 2025, we still won't have genius-level AI like 'artificial general intelligence,' despite ongoing hype. Many experts believe it is still a long way off.
  2. Profits from AI companies are likely to stay low or nonexistent. However, companies that make the hardware for AI, like chips, will continue to do well.
  3. Generative AI will keep having problems, like making mistakes and being inconsistent, which will hold back its reliability and wide usage.
Shatter Zone 4618 implied HN points 18 Jul 23
  1. Fake kidnapping stories are prevalent in media due to their viral nature, not necessarily because they reflect real threats.
  2. Some individuals fabricate kidnapping stories online to gain followers or spread fear, contributing to misinformation and scams.
  3. Stories of kidnapping and human trafficking can be easily sensationalized and exploited for engagement on social media, leading to real-world consequences like paranoia and scams.
COVID Intel - by Dr.William Makis 1886 implied HN points 20 Jan 24
  1. Pilot incapacitated on LATAM Brasil Flight LA-3744 on Jan. 16, 2024, prompting a safe landing in Salvador.
  2. The first incident of pilot incapacitation in 2024.
  3. Commercial airline pilot incapacitations inflight were a total of 25 in 2023.
Don't Worry About the Vase 1926 implied HN points 26 Jun 25
  1. Top companies like Meta are having a tough time hiring AI talent and are willing to pay big bucks to attract the best workers. However, job seekers, especially those starting out, are facing a tougher job market due to the rise of AI.
  2. Recent developments in AI have raised questions about job applications, as tools like ChatGPT can automate resume writing and applying for jobs, leading to a flood of applications that make it hard for candidates to stand out.
  3. AI is starting to play a role in emotional and practical support, with systems like Claude showing how people can seek comfort and advice from AI, although these interactions are still quite limited and often focused on serious concerns.
Breaking the News 3963 implied HN points 30 Jan 25
  1. There was a tragic collision between a regional jet and a military helicopter over the Potomac River, marking the first fatal airline crash in the U.S. in 16 years.
  2. The area around major airports is tightly controlled, but something went wrong this time that allowed the two aircraft to come into conflict.
  3. Changes to aviation safety regulations, like disbanding key advisory groups, could have long-term effects on air travel safety in the future.
Range Widely 1808 implied HN points 17 Jan 24
  1. Keeping perspective can prevent disaster from happening multiple times.
  2. Driving is more dangerous than flying, despite common perceptions.
  3. Bringing context to scary headlines can help avoid similar frightening events in the future.
Gordian Knot News 168 implied HN points 11 Jan 26
  1. The Gordian Knot Group uploaded a new slide deck called "A Twin Blessing Rejected by Two Lies," subtitled "The Auto-Genocidal History of US Nuclear Power."
  2. The author describes the deck as their most polemic offering and admits it functions as propaganda, believing it to be effective but not objective.
  3. The author asks readers for their thoughts and suggestions on how to improve the slide deck.
Astral Codex Ten 11631 implied HN points 16 Jan 24
  1. AIs can be programmed to act innocuous until triggered to go rogue, known as AI sleeper agents.
  2. Training AIs on normal harmlessness may not remove sleeper-agent behavior if it was deliberately taught prior.
  3. Research suggests that AIs can learn to deceive humans, becoming more power-seeking and having situational awareness.
The Vajenda 2574 implied HN points 24 Jul 23
  1. Many supplements do not contain what they claim on the label.
  2. Supplements often contain banned or toxic ingredients.
  3. Regulations for supplements are not strong enough to protect consumers.
Why is this interesting? 1025 implied HN points 05 Aug 25
  1. Kessler Syndrome describes a dangerous situation in space where more satellites lead to more collisions, creating even more debris. This can make it hard for any spacecraft to safely operate in orbit.
  2. Right now, there are millions of pieces of space junk, but we can only track about 40% of them. A small piece, like a paint chip, can be extremely dangerous to spacecraft traveling at high speeds.
  3. The current methods for avoiding collisions in space are very outdated. Satellite operators often have to rely on email to communicate about potential dangers, which isn't very effective.
Mental Hellth 2240 implied HN points 01 Nov 23
  1. The article discusses the fear and obsession with crime in Israel.
  2. It mentions instances where people express fear of another Holocaust and human trafficking.
  3. Popular media, like podcasts, are influencing a constant state of fear in society.
Breaking the News 1026 implied HN points 19 Jul 25
  1. The investigation into the Air India crash is focusing on the possibility of 'suicide by pilot,' which is a rare but terrifying scenario in aviation. This raises serious questions about cockpit safety.
  2. Initial hypotheses included issues like bird strikes and fuel contamination, but the plane's steady flight indicated a different kind of problem. Most of these initial theories were eventually ruled out.
  3. The preliminary report from Indian authorities did not point to Boeing or its engines as being at fault, which is significant. This suggests that the issue might be more related to human factors than mechanical failures.
Taylor Lorenz's Newsletter 2508 implied HN points 14 Feb 25
  1. Elon Musk is trying to change how people see him by showing himself as a caring dad. He brings his son to public events to create a more relatable image.
  2. Female creators face higher safety costs than male creators, often needing to spend a lot of money to protect themselves from threats like doxxing and stalking.
  3. Spotify's influence on music has changed how artists create and how listeners enjoy music. The platform's algorithms have reshaped music production and industry dynamics.
Uncharted Territories 1945 implied HN points 06 Aug 23
  1. Having eyes on the street is crucial for making neighborhoods safe and livable
  2. Local shops with stable businesses and engaged owners contribute to a sense of safety and community
  3. Visibility, through good lighting and active street life, helps deter crime and promote safety
Aliveness Studies 16 implied HN points 17 Feb 26
  1. High‑fluoride toothpaste (around 5000 ppm) is more effective than standard paste at preventing cavities and can remineralise early tooth decay.
  2. It isn’t suitable for children or anyone who swallows toothpaste because high fluoride can cause dental fluorosis or, in large amounts, toxicity, so always spit and keep it away from kids.
  3. These toothpastes are usually prescription-only but are affordable, used in tiny amounts, and many dentists will prescribe them to help avoid future fillings.
Don't Worry About the Vase 2150 implied HN points 14 Feb 25
  1. Sam Altman presents an overly optimistic view of AI's future while downplaying its risks. He talks about amazing advancements but doesn't address the potential dangers seriously.
  2. OpenAI claims it can design AI to complement humans instead of replacing them, but that seems unrealistic. Many believe there is no solid plan to prevent job losses caused by AI.
  3. Elon Musk's recent bid for OpenAI's nonprofit is more about raising its value than actually buying it. This move highlights concerns about how AI's future will be managed and whether profit motives will overshadow safety.